Releases

AllMajorMinorPatch Show pre-releases
1.1.0 minor

@ogygia/tsdown-plugin-svelte

Minor Changes

  • #64 3bff299 Thanks @PuruVJ! - Breaking: features move into plugin packages, each a Vite plugin listed beside ogygia() and before sveltekit() (plugins: [ogygia(), flags(), sveltekit()]; out of order, or without ogygia(), the build says so): flags (@ogygia/flags, was ogygia/flag), frozen pages (@ogygia/freeze, was ogygia/freeze and ogygia({ freeze })), fragment federation (@ogygia/federation, was ogygia/federation; mount moved from ogygia/router) and content (@ogygia/content, was ogygia/content* and ogygia({ content }); ogygia/content/server is @ogygia/content/server, and the svelte config's ogygia.preprocess() / ogygia.extensions() are preprocess() / extensions() from @ogygia/content/vite), each with its factory from <package>/vite. The devtools dock and the profiler move to @ogygia/devtools: ogygia({ devtools, profiler }) becomes devtools({ dock, profiler }), ogygia/profiler* imports become @ogygia/devtools/profiler* (@ogygia/devtools/profiler, /profiler/client, /profiler/storage, /profiler/storage/sqlite|redis|postgres), and the event layer ogygia/devtools exported is ogygia/events (with ogygia/insights beside it). ogygia/testing is @ogygia/testing. The server router is @ogygia/router (was ogygia/router; list router() from @ogygia/router/vite so its pages' CSS links — dev warns when a page renders without it), and its typed fetch client ogygia/router/client is @ogygia/router/api (POST(schema, …) endpoints now type-check, and $infer carries their verbs). An endpoint handler's c.params is now typed strictly by its table key, at the top level, inside layout() tables and behind when(), and its return type reaches api(); GET<'/p'>(fn) is gone (a handler written outside its table checks with satisfies Handler<'/p'>). A plugin is made with definePlugin({ name, server, client, options, build }) from ogygia/vite, which returns a Vite plugin; its server and browser halves are named by location (server: import.meta.resolve('./server.ts')) instead of found by package export name. The command line lives in @ogygia/cli (npx ogygia … keeps working: ogygia's bin runs it). ogygia is now a thin package over @ogygia/core; plugins depend on @ogygia/core. ogygia()'s region options sit at its root: regions.visible, regions.presets, regions.ttl and regions.maxPageAge are visible, presets, ttl and maxPageAge. @ogygia/tsdown-plugin-svelte turns a component's relative ./x.ts imports into ./x.js in the shipped component and its types, so sources can import with real extensions. Fixed: a Kit-hydrated (csr = true) page could link the stylesheets of every region in the app, depending on how the bundler grouped the server build (newer rolldown, under vite 8.3, did).

Patch Changes

  • #64 3bff299 Thanks @PuruVJ! - Component types are emitted with isolatedDeclarations off, so a library that turns it on still gets its .svelte.d.ts; a new tsconfig option names the config to start from.
0.14.0 minor

@ogygia/bake

Minor Changes

  • #1357 55e05d4 Thanks @PuruVJ! - import.meta.og.bake moves to its own package: install @ogygia/bake, list bake() beside ogygia(), and reference @ogygia/bake/types; without the plugin a bake call fails the build with macro_needs_plugin.

Patch Changes

  • #1384 9cb1d28 Thanks @PuruVJ! - In dev, an edit to a module a bake call imports bakes again with the new value, and bakes of one module at once share one run instead of racing on one temp file.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/barrels

Minor Changes

  • #508 eae1add Thanks @PuruVJ! - Breaking: the barrel rewrite is a plugin now, barrels() from @ogygia/barrels, listed beside ogygia() (plugins: [ogygia(), barrels(), sveltekit()]) with the same options. Use it in place of ogygia({ barrels }), which is a build error (barrels_option_removed) that names it; debarrel(), DebarrelOptions and DebarrelMatcher are gone from ogygia/vite.

Patch Changes

  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/cli

Minor Changes

  • #1363 c7ed8cf Thanks @PuruVJ! - The flag inventory (node_modules/.ogygia/flags-manifest.json) is now written by flags(): a build writes it only when flags() is listed beside ogygia(), and an app that imports @ogygia/flags without listing it gets a one-time plugin_not_listed build warning.
  • #528 9201d45 Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking: the internal internal/compiler-browser and internal/compiler-browser-full import paths are gone. The compiler runs in Node only; use @ogygia/core/internal/compiler there. - @ogygia/cli: breaking: the ogygia mcp server no longer has the ogygia_observatory tool; the Observatory it linked to is gone. - @ogygia/content: its macros and markdown pipeline read the compiler through @ogygia/core/internal/compiler and Node's own fs, path and crypto.

Patch Changes

  • #1192 b8d8cf0 Thanks @PuruVJ! - The CLI's perf measure and the MCP profile tool ask the profiler's /page and /reset with a POST, as those routes now take a POST only.
  • #475 81a40ea Thanks @PuruVJ! - Durations are measured on the monotonic clock: the memory store's TTLs, a federation deploy check's retry wait, the profiler's recording window, deadlines and busy waits, the testing API's wait, and the perf CLI's waits. A wall-clock step no longer moves them.
  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #1398 694ea84 Thanks @PuruVJ! - Text taken from markup keeps the words the browser shows: the site search index no longer drops prose between a < and a later > (a < b and c > d), indexes no script or style text, and keeps inline code words; the MCP profile tool quotes an HTML error page by its words, not its CSS or script.
  • #875 0160249 Thanks @PuruVJ! - ogygia mcp checks each tool's arguments against the tool's declared schema and says which one is wrong, instead of failing with "Internal error" or silently using a default.
  • #1815 8db8d12 Thanks @PuruVJ! - ogygia_debug lists the features the runtime booted in the story header, instead of always saying "installers ran".
  • #1815 8db8d12 Thanks @PuruVJ! - ogygia_debug shows why a region's hydration failed, in its block and in the warnings, instead of only "HYDRATION FAILED".
  • #877 152a2bd Thanks @PuruVJ! - ogygia mcp answers every request exactly once: a malformed one gets a JSON-RPC error instead of no answer (which left the client waiting).
  • #895 adcd63f Thanks @PuruVJ! - ogygia_profile says a locked profiler is locked (and asks for its secret, or says the key was rejected) instead of failing with "profiler did not return JSON".
  • #1910 bd83130 Thanks @PuruVJ! - The MCP debug story shows each applied deferred hole in a block of its own, keyed by its endpoint: its HTML size, how long its fallback showed, and how many times a live hole re-applied. It no longer has a "props crossed the wire" line, which nothing ever fed.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/compiler

Minor Changes

  • #1357 55e05d4 Thanks @PuruVJ! - import.meta.og.bake moves to its own package: install @ogygia/bake, list bake() beside ogygia(), and reference @ogygia/bake/types; without the plugin a bake call fails the build with macro_needs_plugin.
  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #1665 096de6c Thanks @PuruVJ! - Breaking: shipping only the page.data your islands read is a plugin now, dataShake() from @ogygia/data-shake/vite, listed beside ogygia() (plugins: [ogygia(), dataShake(), sveltekit()]). Without it every island that reads the page gets all of page.data, and every server hole looks the page up.
  • #1363 c7ed8cf Thanks @PuruVJ! - The flag inventory (node_modules/.ogygia/flags-manifest.json) is now written by flags(): a build writes it only when flags() is listed beside ogygia(), and an app that imports @ogygia/flags without listing it gets a one-time plugin_not_listed build warning.
  • #1705 78c6d48 Thanks @PuruVJ! - wake on a render: 'deferred' import or preset, deprecated in 0.13, now fails the build (hole_wake_removed): a deferred region ships no JS, so wake never applies. Write its timing as fetch: with { render: 'deferred', fetch: 'visible' }.
  • #1354 53e96f0 Thanks @PuruVJ! - A named import takes the region mark, in .svelte, .ts and .js, and replaces import.meta.og.asRegion, which is now a build error (as_region_removed) that prints the import to write. Before: import { Search } from '@ogygia/content'; const SearchBox = import.meta.og.asRegion(Search, { wake: 'idle' }); After: import { Search as SearchBox } from '@ogygia/content' with { wake: 'idle' };. One with covers every name in its import; a marked name used as anything but a component, and a mark on a namespace, type-only or bare import, fail the build with a code.
  • #1678 ea6effb Thanks @PuruVJ! - The deprecated ogygia.extensions() and ogygia.preprocess() from ogygia/vite are removed (they did nothing): drop them from svelte.config.js, or for markdown use extensions() and preprocess() from @ogygia/content/vite.
  • #1704 ddbf46e Thanks @PuruVJ! - An import attribute ogygia does not know on a .svelte component import (a misspelling like wkae, or the retired partial) now fails the build as region_attribute_unknown, naming the key and listing the ones the app has; the partial attribute's own error is gone. Write region: 'raw' for a held component, or wake: … for an island. type and attributes on other imports are left alone.
  • #1665 096de6c Thanks @PuruVJ! - Breaking: postrender() requires dataShake(). Add import { dataShake } from '@ogygia/data-shake/vite' and list dataShake() beside postrender(). Without it the build fails with plugin_needs_plugin, because every deferred region would look the page up and render the stored page again.
  • #1704 ddbf46e Thanks @PuruVJ! - The retired ogygia() option keys (regions, regionTtl, continuity, preload, plugins, freeze, content, devtools, profiler) no longer have their own messages: any key ogygia() does not have, a misspelling included, now fails the build as option_unknown and lists the keys it has. Move a retired key to its replacement (ttl, router: { forms }, or the feature's own plugin beside ogygia()).
  • #1688 0602ba2 Thanks @PuruVJ! - The build no longer appends a registration call after each exported createContext(...), and the no-op __tag_context it called is gone from ogygia and ogygia/internal: a module that exports only contexts is left exactly as written. Rebuild any package that shipped already-transformed output.

Patch Changes

  • #1323 e1f6a87 Thanks @PuruVJ! - A barrel module with an import kept only for what loading it does (import './theme.css', a polyfill, a namespace import it never exports) is no longer taken for a pure barrel, so the barrels pass keeps importing it and the CSS or polyfill still loads.
  • #1640 13b7427 Thanks @PuruVJ! - A portable snippet or import.meta.og.$ closure now captures a host name read after a class whose static {} block declares the same name, and no longer captures a class expression's own name read inside it.
  • #1661 6c215fd Thanks @PuruVJ! - A browser build that loads one of ogygia's server modules by its resolved id now fails with virtual_wrong_realm, instead of getting the module's browser stand-in.
  • #1413 938f79f Thanks @PuruVJ! - On Kit 2 without OGYGIA_SECRET, the build's check that no browser file holds the signing key now looks for the key the server bundle really uses (before, it looked for a key nothing used, so that half of the check never fired) (#1412).
  • #1520 6cfab1a Thanks @PuruVJ! - A module the compiler rewrites in several passes (a macro, the island transform, the module-script registrations, the client $app/* shim, the server's ownership stamps) now carries a source map back to the file as written, so stack traces and breakpoints land on the right line and column. A text-only plugin macro leaves a map with no positions rather than wrong ones.
  • #1188 cc1b757 Thanks @PuruVJ! - A second client build in one process (vite build --watch) emits every island entry again and writes full location maps: it emitted none, so the rebuilt pages could not load their islands.
  • #1431 cf4be5b Thanks @PuruVJ! - A component's script and style blocks are read the way its markup is read: a <script> or <style> inside a comment or another block's text is no longer taken as code, a > inside an attribute value no longer cuts the tag (flag sites were missed), data-x="module" no longer marks the module script, and the CSS preprocess keeps a TypeScript script's $$props as written (#1428).
  • #1541 1a475de Thanks @PuruVJ! - A context that cannot cross into regions is now said, not dropped silently: a one-off Symbol() key (context_key_one_off) or a function value (context_value_function) set with setContext from ogygia or in a <Provide> warns in dev, naming the key and what to use instead; the build warns on a one-off Symbol() key it can see (region_context_key_one_off), and its "island code reads a context nothing bridges" warning now covers Symbol.for(name) keys.
  • #1295 2fea150 Thanks @PuruVJ! - A deferred hole, a lake, a live region or a held region written directly in a csr=true +page.svelte or +layout.svelte now stays a real region, as the same import already did in a shared component. Before, it became a plain component: the hole rendered inline in the page's SSR and the lake was hydrated by Kit. Wake islands in a csr=true route file are still plain components Kit hydrates.
  • #1546 a4de38a Thanks @PuruVJ! - In dev, a page whose csr changes while the server runs (an option file added, edited or removed above it) is compiled again for its new csr, so a page that becomes csr = false gets its regions without a restart (before, its marks stayed stripped and its regions never hydrated) (#1375).
  • #1972 c688e01 Thanks @PuruVJ! - On the dev server, an og.$ function that a page hands an island (through context or props) now works. Before, the dev server sent no function table with the page, so the island's call threw ... is not a function. Builds were not affected.
  • #1546 a4de38a Thanks @PuruVJ! - In dev, a region added while the server runs reloads the runtime entry and every other module that read the regions, so its wake works without a restart (before, a page's first idle or visible region never woke until a restart) (#1376).
  • #1546 a4de38a Thanks @PuruVJ! - In dev, every route file Kit reads counts when it is added, edited or removed (a +page.md page too, not only .svelte/.js/.ts), and an edit that changes no route's options (a page's body, a comment in +layout.ts) no longer reloads the route modules (#1495).
  • #1423 6fcf4e2 Thanks @PuruVJ! - In dev, a server-router page component's CSS is compiled as the build compiles it: <style lang="scss"> and <script lang="ts"> are preprocessed first, so its styles are scoped in dev as they are in production (#1422).
  • #1546 a4de38a Thanks @PuruVJ! - In dev, deleting the app's last route file no longer throws ENOENT ... scandir src/routes: a routes directory that is not there holds no routes (#1496).
  • #1546 a4de38a Thanks @PuruVJ! - In dev, a wire class, a .remote.ts file or a router page module added or edited while the server runs takes effect without a restart: the wire loaders, Kit's codecs and the router CSS follow it (before, they kept what the server saw at start) (#1395).
  • #1381 e42f8d0 Thanks @PuruVJ! - import.meta.og.$ captures exactly what its function reads from outside: a name a block or an inner function shadows is still captured where it is not shadowed, an imported name that shares a global's name (fetch) is captured instead of read as the global, and a class method name or import.meta is no capture; a $() inside another $() fails the build.
  • #1558 d3f9cc7 Thanks @PuruVJ! - A function marked with import.meta.og.$ in TypeScript hoists as plain JS with optional parameters, ! assignments, local types and interfaces, <T>x casts, this parameters and class modifiers; an enum, a namespace with values or a parameter property inside it fails the build at its line instead of breaking the page.
  • #1181 8fa29ec Thanks @PuruVJ! - A dynamic import() that names a region key is now refused with dynamic_region_import in a .ts / .js module too, and for the region key as well. Before, it was left in place and did nothing.
  • #1321 a46af31 Thanks @PuruVJ! - Two modules with side effects that export * each other no longer hang the dev server or the build (it ran out of memory); the export walk visits each module once, as JavaScript does.
  • #1899 1f2bdb7 Thanks @PuruVJ! - A build no longer prints relocation_unpatched for the fn manifest. The browser bundle carried a leftover og.$ registration module whose contents the bundler always dropped; the page's inline manifest is what delivers the factories, so the module is gone.
  • #1392 586d8d6 Thanks @PuruVJ! - The og.$ fn-manifest is written sorted by tag, so two builds of the same app give the same runtime chunk whatever order the bundler transformed modules in.
  • #1970 f9ebb02 Thanks @PuruVJ! - An island from a mounted peer page now reads the cross-island context its page provides (<Provide>, setContext, createContext), even when the shell provides no context of its own. Before, a production shell left the context bridge out and never handed context to a peer's island, so the island hydrated with its context silently empty. A federated build now also ships the live-region feature a peer's held or kept regions need.
  • #1248 fe5c2c8 Thanks @PuruVJ! - A {#snippet} that crosses into an island now brings along the functions its use:, transition:, in:, out: and animate: directives name, the values its style: directives read, <svelte:component this={…}>'s component, and a host name that a for loop only shadows. Before, the snippet's entry referenced them undeclared and failed at render.
  • #1374 5c5b673 Thanks @PuruVJ! - A portable snippet now carries the names its pattern defaults read ({#each xs as { a = dflt }}, a snippet parameter's default, {@const}, {#await … then}), its {@debug} names, and a name read after a catch clause that bound the same name; each was missed and threw a ReferenceError on the far side, or left the snippet native.
  • #1316 55177a4 Thanks @PuruVJ! - A deferred hole whose server tree reads the page through a $lib re-export of page, getStores() or import('$app/state') now gets page.data; before, it rendered with an empty one.
  • #1203 2378a3c Thanks @PuruVJ! - A production server that cannot find or read the client build's handoff (og-region-deps.json) now says so (island_deps_handoff_missing, with the paths it tried) instead of rendering every hole without its CSS and preloads in silence. A handoff that failed to read or parse is tried again on the next request instead of being replaced by an empty one for the life of the process.
  • #1691 e39d03c Thanks @PuruVJ! - An app's Kit config (kit.files.routes, kit.outDir, kit.inlineStyleThreshold) is now held by the plugin that read it, so it can never fall back to the defaults however many apps one process configures.
  • #1906 9d5cdfa Thanks @PuruVJ! - ogygia:server/kit-wire re-exports only SvelteKit's create_remote_key, the one helper ogygia uses, which Kit 2 and Kit 3 share; a canary checks it in every installed Kit.
  • #1731 b60d52f Thanks @PuruVJ! - In dev, a region module (a lake's wrapper) that was loaded while no file held it, after a missed file-watcher event, is reloaded once a later save gives it a host, instead of staying empty until a restart.
  • #1477 5a43fea Thanks @PuruVJ! - A deferred or held region in a library component only the server bundle reaches gets its styles built: the client leg takes the server leg's final list of server regions, so the hole no longer shows unstyled.
  • #1641 e61ca82 Thanks @PuruVJ! - A portable snippet now captures a host name read by a legacy let: directive's default (let:item={[a = d]}) instead of failing with a ReferenceError when it renders.
  • #1347 0ecb8af Thanks @PuruVJ! - An import.meta.og.code, md, loader or regions call in an app without content() listed now fails the build with macro_needs_plugin, naming @ogygia/content; before, it reached the browser or the server untouched and threw a bare TypeError there.
  • #1456 78f3edd Thanks @PuruVJ! - An import.meta.og.* call after a regex that follows a keyword (return /'/.test(s)) or an if (…), or inside a template's ${…}, is found and rewritten: the macro scanner lost its place there and left the call in the module.
  • #1426 cea7fe0 Thanks @PuruVJ! - A context provider or a region reached through a local barrel (export { setContext } from 'ogygia', export * from 'ogygia'), a namespace handed on, or a dynamic import('ogygia') now ships the context bridge and Kit's region codec: before, islands lost that context and Kit could not decode the region (#1424).
  • #1230 b168cf7 Thanks @PuruVJ! - Seed shaping now ships all of page.data when the page is put in an array (other than derived([…], fn)'s sources) or a method is called on page.data (page.data.hasOwnProperty('x')). Before, both were read as a narrower key set, so an island could read undefined for a key that was left out of the seed.
  • #1237 13b589f Thanks @PuruVJ! - Seed shaping now covers what a function handed the page reads through an alias of its parameter, a default parameter or arguments, and it no longer follows a function binding that is later assigned another function. Before, those keys could be left out of the shipped page seed.
  • #1256 4b7d56b Thanks @PuruVJ! - Seed shaping ships all of page.data when a module re-exports the page store (export { page } from '$app/state', export *) or loads it with import(), so an island reading the page through such a barrel no longer reads undefined.
  • #1319 886446f Thanks @PuruVJ! - Seed shaping ships all of page.data to an island that reads the page through getStores() or a parameter default (function f(p = page)), and no longer mistakes a shadowed helper name or a local window for the module's helper or the browser global; before, those reads could be left out of the page seed.
  • #1258 5c3b43c Thanks @PuruVJ! - Seed shaping treats get and derived as Svelte's store helpers only when they are imported from svelte/store (under any name), and a derived(page, fn) whose callback is not written inline ships all of page.data. Before, a get(page, 'data.x') from another library or a foreign derived could leave keys out of the page seed.
  • #1521 b970e6d Thanks @PuruVJ! - A region preset takes the standard keys (render, wake, fetch, …) when the app renames its import keys, as the docs say; before, a preset with render failed the build under importKeys: { render: 'mode' }, and the unknown-key message left out keep.
  • #1686 07d2165 Thanks @PuruVJ! - The dev server sees a file change even after more than 100,000 cached compiler answers: the cache no longer drops what a file read, only answers computed from it.
  • #1240 bae4982 Thanks @PuruVJ! - A route's csr option is now read with a parser, the same way Kit reads it. export let / export var, a declaration with several declarators, export { local as csr } and a parenthesized value now count, and text in a string or comment no longer does. Before, such a page could be taken for the wrong kind: its islands stripped on a csr = false page, or the client build predicted wrong.
  • #1359 5ff44f8 Thanks @PuruVJ! - A region that a page and its layout both import now names the same host in every build, whatever order the bundler transforms them in.
  • #1419 88858cc Thanks @PuruVJ! - A component marked as a region is one region however it is imported: $lib/…, a relative path, a root-absolute /src/… path or an app alias give one id, and a host with <script lang="ts"> shares the same wrapper as a plain one (before, a root-absolute or aliased import made a second region, unstyled in dev, and the two hosts' wrappers overwrote each other) (#1415).
  • #1371 1073b6a Thanks @PuruVJ! - A region import mark that would do nothing now fails the build: render: 'static' alone, keep on a hole, lake or live region, and a call-shaped schedule like wake: 'visible(200px)' (a media query is now media words around balanced conditions, everywhere a schedule is read).
  • #1303 709372e Thanks @PuruVJ! - A route that sets a claimed config key in both its universal and its server module now takes the server one, as SvelteKit does at run time (config merges the other way from csr).
  • #1282 8bc2778 Thanks @PuruVJ! - A route node that sets csr in both its universal and its server option file now takes the universal one, as SvelteKit does. Before, the server one won for page worlds, so a page Kit hydrates could be treated as client-off.
  • #1298 d70e275 Thanks @PuruVJ! - A page or layout with a layout reset ([email protected], +layout@(group).svelte) is now a route to ogygia, read the way SvelteKit reads it: its csr and its route config come from the layouts it actually renders in, and a dev edit to it refreshes the route answers.
  • #1455 1b08deb Thanks @PuruVJ! - A build warns (runtime_marks_late) when a file it first reaches while bundling places a region that needs a runtime feature the build already left out, naming the file and the feature (#1452).
  • #1389 493898d Thanks @PuruVJ! - A second build in one process (vite build --watch, or build() called twice) scans the app again and reads each route's csr afresh, so a region added or a csr flipped between builds is in the next build.
  • #1047 acfd819 Thanks @PuruVJ! - The client build also fails when it holds a signing secret written as an escaped string literal (a secret with a quote, backslash, newline, slash or non-ASCII character), not only the raw text.
  • #1421 dfb60d0 Thanks @PuruVJ! - A build lists its server regions by id in the values it writes at the end of the server leg (the manifest's maps, the late loaders, the page-data reads), so two builds of one app give the same server chunk bytes whatever order the bundler reached their hosts in.
  • #1421 dfb60d0 Thanks @PuruVJ! - A deferred or held region in a file the build only reaches late (a library outside src/ with no ogygia.files) is now served in production: the build fills in the server manifest from its final list of regions and gives a late region's loader its own chunk. Before, the region was missing from the manifest and its requests were refused (#1420).
  • #1366 4fda955 Thanks @PuruVJ! - import.meta.og.source in a .svelte <script module> puts its runtime import inside the script (it was put in front of the whole file, as page text), and adds no line to any module, so error lines stay right.
  • #1487 c429240 Thanks @PuruVJ! - A source file the build's first scan cannot parse is now treated as one that may define a wire class, so a wire class in it keeps its loader and the wire feature; before, it could fail to decode on the client with no message.
  • #1268 234a23d Thanks @PuruVJ! - A region import in a .ts / .js module is read from the module's syntax tree: a comment or line break before with no longer hides it, and a named, namespace, bare or re-export form with a region key is refused (region_import_invalid) instead of passing through as a plain import.
  • #1389 493898d Thanks @PuruVJ! - A watch rebuild marks every file an island reaches as island code again, so a shared module keeps ogygia's $app/* shim on the rebuild too (#1448).
  • #1313 95dc4e3 Thanks @PuruVJ! - A wire class whose codec member is typed (static wire: Codec<Cart> = import.meta.og.wire(…)) or readonly is now found by the build, so it gets its loader and the wire feature; before, it failed to decode in production. A module with only an unrelated computed static, or a wire class it does not export, is no longer imported into every realm.
  • #1311 779fa59 Thanks @PuruVJ! - An island whose props carry a live value the build did not predict (a forwarded children, a snippet passed as a prop, a spread of $props()) now hydrates: the runtime loads the wire codec when that island wakes, instead of failing with "wired prop but no wire feature". Builds that already ship the codec are unchanged.
  • #1170 c92ebbb Thanks @PuruVJ! - A wire class exported with export default (export default class Cart, or class Cart {} then export default Cart) is now registered, tagged #default, and an anonymous export default class with a wire codec is the build error wire_class_anonymous. Before, the first instance of either to cross into an island threw transportable_unregistered.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/content

Minor Changes

  • #1677 2a4998f Thanks @PuruVJ! - The deprecated ContentEntry type is removed: use ContentRef (a ref, without a body) or Entry (what get() returns, with its body).
  • #528 9201d45 Thanks @PuruVJ! - - @ogygia/content: breaking: snippet_macros and its SnippetMacro type are no longer exported from @ogygia/content/markdown; only the removed Observatory used them, and nothing replaces them.
  • #1842 b6b930e Thanks @PuruVJ! - An entry's body (and a format's and a source's) is typed as the AwaitableRegion it is, so await entry.body type-checks without a cast.
  • #600 123cc08 Thanks @PuruVJ! - breaking: frontmatter tags: [svelte] is now a one-item list (['svelte']), as YAML reads a flow sequence; it used to be the string '[svelte]'. A value that should stay a bracketed string (title: [create your own]) needs quotes: title: '[create your own]'. A value that only starts with brackets ([docs](/docs)) stays text.
  • #601 a7700e8 Thanks @PuruVJ! - breaking: import.meta.og.md() with a component tag in its prose (<Foo />, <ui.Button />) is now the build error content_md_dynamic; it used to render the tag as an inert unknown element. Use a component for the live part, or put the tag in inline code or a fence to show it as text.
  • #1842 b6b930e Thanks @PuruVJ! - The remark and rehype markdown options are typed: each entry is a plugin, a [plugin, ...options] tuple, a unified preset ({ plugins, settings }) or a staged { enforce, plugin }, where they were any before.

Patch Changes

  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #1225 6e91a78 Thanks @PuruVJ! - Backlinks follow a collection replaced by a dev reload: the index is kept only while it was built from the same collections at their current versions. A replacement with the same size and version could serve the old collection's backlinks.
  • #655 23c0797 Thanks @PuruVJ! - BlogList and BlogPost show a post's date as the day it names; a server west of UTC no longer shows the day before.
  • #654 8ebdb75 Thanks @PuruVJ! - BlogList with group={false} lists the posts in the order given, as documented, instead of sorting them by date.
  • #1466 e451135 Thanks @PuruVJ! - content({ markdown }) now says content_config_read_first when the svelte config was loaded before it and so read no markdown settings, instead of leaving every markdown file uncompiled with no message.
  • #646 1fdb732 Thanks @PuruVJ! - Doc's keepReading prop now sets the "Keep reading" cards, as its docs say; before, the cards always showed the trail's suggested pages.
  • #651 f93094a Thanks @PuruVJ! - Frame no longer asks the site for its nav tree when its sidebar is replaced or removed.
  • #702 f5fecc9 Thanks @PuruVJ! - @ogygia/content: a frontmatter key named __proto__ is kept as a plain key of the parsed object, as JSON.parse does, and never changes the object's prototype.
  • #1891 950a623 Thanks @PuruVJ! - A kept search and sidebar follow a new base after a navigation: the search link and palette go to the new base's search page, and the sidebar saves its scroll under the new base.
  • #1379 ccf170e Thanks @PuruVJ! - A content page's module script is found as Svelte reads its tag: one with a > inside an attribute value is still the module (no second module script is added), and an instance script whose value only says module is not taken for it.
  • #1154 2779907 Thanks @PuruVJ! - The docs shell's mobile bar is kept across client navigations, and a link picked in its sheet closes the sheet (it stayed open over the new page).
  • #441 352a522 Thanks @PuruVJ! - A code fence's language, permalink id and file name are escaped into their HTML attributes: an info string can no longer put markup into the page.
  • #556 db8271e Thanks @PuruVJ! - parseFrontmatter reports a frontmatter block with an empty key or a tab indent as content_frontmatter_invalid printed once (one [ogygia], one link), with the parser's words as the reason, instead of nesting a second prefix and link inside it.
  • #622 a9bf48a Thanks @PuruVJ! - A frontmatter list item with a [ or { inside it (title: [see [a, b]]) is now the content_frontmatter_invalid error, as YAML reads it, instead of a list split in the wrong place (['see [a', 'b']).
  • #555 ef3bd78 Thanks @PuruVJ! - A live.list stream no longer sends a stale first list and then misses an edit that landed while that list was being read.
  • #1398 694ea84 Thanks @PuruVJ! - Text taken from markup keeps the words the browser shows: the site search index no longer drops prose between a < and a later > (a < b and c > d), indexes no script or style text, and keeps inline code words; the MCP profile tool quotes an HTML error page by its words, not its CSS or script.
  • #548 5900557 Thanks @PuruVJ! - An outline read before its collection has loaded is woven once, not again on the next read.
  • #528 9201d45 Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking: the internal internal/compiler-browser and internal/compiler-browser-full import paths are gone. The compiler runs in Node only; use @ogygia/core/internal/compiler there. - @ogygia/cli: breaking: the ogygia mcp server no longer has the ogygia_observatory tool; the Observatory it linked to is gone. - @ogygia/content: its macros and markdown pipeline read the compiler through @ogygia/core/internal/compiler and Node's own fs, path and crypto.
  • #865 00720a1 Thanks @PuruVJ! - A search hit's excerpt marks the query's terms in one pass, so a term never lands inside another term's <mark> or an HTML entity and breaks the excerpt's markup.
  • #599 bd66220 Thanks @PuruVJ! - site(): a context function that throws is now reported as content_site_context_failed (with the thrown error) instead of being swallowed. The request still reads the public pages only, as before.
  • #441 352a522 Thanks @PuruVJ! - A tab label, tab group name or ::: container title holding { or } now stays text, instead of being compiled by Svelte as an expression.
  • #1149 9829521 Thanks @PuruVJ! - @ogygia/content: a ::: code-group fence whose info starts with the fence character (~~~ ~x [t]) no longer swallows the tab wrappers after it, and a line whose info holds a backtick is no fence inside ::: tabs / ::: code-group`, as in CommonMark.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/core

Minor Changes

  • #1680 9fb9b44 Thanks @PuruVJ! - import.meta.og.asRegion is gone from the import.meta.og types too, so editors flag a leftover call before the build does: mark the import itself instead (import { Header as H } from '@design/system' with { wake: 'load' }).
  • #1357 55e05d4 Thanks @PuruVJ! - import.meta.og.bake moves to its own package: install @ogygia/bake, list bake() beside ogygia(), and reference @ogygia/bake/types; without the plugin a bake call fails the build with macro_needs_plugin.
  • #508 eae1add Thanks @PuruVJ! - Breaking: the barrel rewrite is a plugin now, barrels() from @ogygia/barrels, listed beside ogygia() (plugins: [ogygia(), barrels(), sveltekit()]) with the same options. Use it in place of ogygia({ barrels }), which is a build error (barrels_option_removed) that names it; debarrel(), DebarrelOptions and DebarrelMatcher are gone from ogygia/vite.
  • #532 b011a71 Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking (internal paths only): ogygia/internal/region-core is gone, and ogygia/internal/compiler no longer exports set_parser or takes a pathModule option; they existed only to run the compiler in a browser. Import region and isRegion from ogygia. Build output is unchanged.
  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #1684 22e276e Thanks @PuruVJ! - createContext's old callable form is removed: no theme('dark') <Provide> entry, no theme.get(), no theme.key and no default argument. Destructure const [getTheme, setTheme, hasTheme] = createContext<Theme>('theme'), call the setter above, and ask hasTheme() where you used a default.
  • #1541 1a475de Thanks @PuruVJ! - createContext<T>(name) from ogygia now returns Svelte's [get, set, has] shape, keyed by Symbol.for(name) so the value reaches every region in the browser (const [getRuntime, setRuntime] = createContext<AppRuntime>('app-runtime')).
  • #1878 6ebfbd9 Thanks @PuruVJ! - A bare snippet now freezes through a tiny component ({@render snippet()}) rendered by Svelte's public render(), instead of a hand-written copy of Svelte's private server component shape. Frozen snippet HTML gains one empty marker (<!---->) at its end, so frozen snippets' ids, island fingerprints, freeze keys and ETags change once on upgrade.
  • #1870 97bec00 Thanks @PuruVJ! - In an app without Svelte's async mode, a <Region of={promise}> whose promise rejects now sends the error to the nearest <svelte:boundary>, as async mode does. Before, the error was an unhandled rejection no boundary could catch. With no boundary, the page still gets the promise's own error as an unhandled rejection; a region that left the page before its promise failed reports nothing.
  • #527 a4f938a Thanks @PuruVJ! - - @ogygia/insights: new package, the analysis engine the devtools dock, the profiler's report and @ogygia/testing share (what the browser measured, turned into findings). - @ogygia/core: breaking: the @ogygia/core/insights import path is gone. Install @ogygia/insights and import from it instead; the names are the same.
  • #762 99ceeb4 Thanks @PuruVJ! - core: the import.meta.og.* macros, bake above all, work only in the app's own source and in files a package lists in ogygia.files; an installed file outside that list that uses one fails the build with macro_outside_app, and its bake function never runs
  • #1354 53e96f0 Thanks @PuruVJ! - A named import takes the region mark, in .svelte, .ts and .js, and replaces import.meta.og.asRegion, which is now a build error (as_region_removed) that prints the import to write. Before: import { Search } from '@ogygia/content'; const SearchBox = import.meta.og.asRegion(Search, { wake: 'idle' }); After: import { Search as SearchBox } from '@ogygia/content' with { wake: 'idle' };. One with covers every name in its import; a marked name used as anything but a component, and a mark on a namespace, type-only or bare import, fail the build with a code.
  • #1838 9c6692d Thanks @PuruVJ! - During an upgrade from 0.12.x, an old page's deferred regions are now refused by the new server, so they keep their fallback until the visitor reloads the page. Before, the new server answered them and they hydrated wrong (the old runtime loaded the new build's island and it failed). The refusal is x-ogygia-deny: shape for any region URL minted before capability v3: a single request, a renewal and a batch alike.
  • #693 b0a4f0e Thanks @PuruVJ! - region(C, props, { version }) in @ogygia/core: when two answers for the same held region reach the browser (two commands that both return it), the one with the newer data version stays on screen, whichever arrived last.
  • #1908 f725fef Thanks @PuruVJ! - A new client hook, reloadNeeded({ url, error, regions }), fires once per page when a chunk can load only after a reload, at the same moment the page says chunk_reload_needed. This happens when a download it needs failed (a dropped connection mid-load) and the browser keeps that failure for the page's life. regions are the regions whose own file the chunk is, so an app can offer the reload. A page with no subscriber carries none of its code.
  • #711 41f5eec Thanks @PuruVJ! - breaking: a remote function's argument sent from the browser must be plain data. The server now refuses an argument that holds an ogygia value (a region, a snippet, a function, a wired class instance, a store or a derived store) with a 400 and remote_arg_not_data, and runs nothing. Values the server sends to the browser are unchanged.
  • #528 9201d45 Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking: the internal internal/compiler-browser and internal/compiler-browser-full import paths are gone. The compiler runs in Node only; use @ogygia/core/internal/compiler there. - @ogygia/cli: breaking: the ogygia mcp server no longer has the ogygia_observatory tool; the Observatory it linked to is gone. - @ogygia/content: its macros and markdown pipeline read the compiler through @ogygia/core/internal/compiler and Node's own fs, path and crypto.
  • #1707 b20ae04 Thanks @PuruVJ! - - breaking: an app's own pass over the HTML ogygia sends is the new @ogygia/rewrite package. ogygia.handle({ transform }) now throws handle_transform_moved; list rewrite() (from @ogygia/rewrite/vite, no options) beside ogygia(), and put a rewrite() handle before ogygia.handle() in sequence(). ogygia/markup is removed, and scanRegions / liftRegions / restoreRegions leave ogygia/server. - @ogygia/rewrite: rewrite({ match, render }) renders only the outermost custom elements match picks (a tag prefix, a list of tags, or a test), in one render(elements, ctx) call per chunk of the page and per region answer, so a streamed page keeps streaming and its later chunks are rendered too. - @ogygia/rewrite: each element render gets has ogygia's marks numbered from 0, so a cache can key on its bytes; equal elements in one call are rendered once. - @ogygia/rewrite: ctx carries kind, csr, url, event and signal, an AbortSignal that fires when the request is cancelled. - @ogygia/rewrite: a render's sheets go out once per key per response, into the head while it is open and inline before the first element that needs them after. - @ogygia/rewrite: a failed batch is retried one element at a time; an element that still fails, drops ogygia's marks, or returns HTML that does not close what it opens goes out as ogygia rendered it, with a warning in every build (rewrite_element_failed, rewrite_marks_lost), never thrown. - @ogygia/rewrite: an element split across streamed chunks goes out unrendered, said in dev (rewrite_element_split). - @ogygia/rewrite: several rewrites in sequence() run in order on one marking of the HTML; one listed after ogygia.handle() throws rewrite_after_ogygia, and a rewrite without the plugin throws rewrite_plugin_missing. - @ogygia/rewrite: rewrite({ page }) rewrites the whole document (held until its last chunk) or region answer, says rewrite_page_failed when it throws (thrown in dev), and has .html(str) for a plain string. - @ogygia/rewrite: .check(html, { kind, csr }) runs the whole round trip on one piece of markup for a test, listing failures and, with a DOM, every restored element that differs from Svelte's. - @ogygia/rewrite: a page where nothing was restored ships no restorer, and each streamed chunk that restored something ends with one short restore call. - @ogygia/devtools: the profiler reads a page's regions itself and no longer depends on @ogygia/rewrite.
  • #516 368b1da Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking: SharedState is no longer exported from ogygia or @ogygia/core. Import it from @ogygia/federation/shared instead; it works the same. - @ogygia/federation: SharedState (state shared across separate builds on one page) is exported from @ogygia/federation/shared.
  • #1688 0602ba2 Thanks @PuruVJ! - The build no longer appends a registration call after each exported createContext(...), and the no-op __tag_context it called is gone from ogygia and ogygia/internal: a module that exports only contexts is left exactly as written. Rebuild any package that shipped already-transformed output.
  • #1842 b6b930e Thanks @PuruVJ! - These methods are now typed as function properties, so passing one on by itself (const { handle } = router) type-checks under unbound-method; none of them used this. A method type was bivariant in its parameters and a function property is checked strictly, so code that assigned a narrower function to one of them may now need its parameter types widened. The changed types: ogygiaTransport.Region.encode / .decode and the ref entry's encode / decode (@ogygia/core), CapabilitySigner.sign / .verify (@ogygia/core), Router.handle (@ogygia/router), Signer.sign / .verify and Staging.discard (@ogygia/machine).

Patch Changes

  • #1848 558b9a9 Thanks @PuruVJ! - On SvelteKit 3, a value of a custom transport type now ships in the page seed, the remote seed and a remote call's page facts; before, each dropped it, because ogygia read the transport from Kit's private request state, which Kit 3 no longer keeps it on.
  • #1387 d16b921 Thanks @PuruVJ! - An island whose hydration changed only attributes is read as the browser reads its tags: a > inside a quoted value no longer cuts the tag, values in any quoting with spaces around = and names in any case are compared, and the first of two same-name attributes is the one kept.
  • #1121 d61549f Thanks @PuruVJ! - @ogygia/core: Tag and attribute names read from server HTML are lower-cased the way a browser does it, ASCII letters only, so data-Äsrc and data-äsrc count as two attributes. @ogygia/federation: A peer tag with two attributes whose names differ only past ASCII (data-Äsrc, data-äsrc) now has each value pinned in its own attribute. Before, the first value was written into the second.
  • #1845 c61a561 Thanks @PuruVJ! - A wake: 'load' island whose background start throws is reported (machine_queue_start_failed, then the error to the page's error handler), and the starts queued behind it still run in the same background task instead of waiting for another.
  • #1841 6e90f0c Thanks @PuruVJ! - A held region the server baked with its stylesheet keeps its styles when it repaints with the same sheet (an island that reads the region's query again once it hydrates): the answer's sheet is now linked in the head whenever the head has none, not skipped because the old copy inside the region was still there.
  • #509 936654e Thanks @PuruVJ! - - @ogygia/core: the dock's island-barrel finding says why barrels() left the barrel (it runs code of its own, it is a package the plugin does not list, an import of it marks a region, and so on) with the fix for that reason, and names barrels() from @ogygia/barrels when the app does not list it. - @ogygia/devtools: the profiler's island-barrel warning quotes the same reason, recorded by the build, and its fix names barrels() instead of the removed ogygia({ barrels }).
  • #1161 8644a71 Thanks @PuruVJ! - A hole batch called off while its body is still arriving (two quick link clicks), or whose body breaks off, now ends quietly: no more "Uncaught (in promise) AbortError" for error monitors to report. Its holes still fetch on their own.
  • #1216 0b97707 Thanks @PuruVJ! - Holes fetched in one batch (a client-side navigation) mint their slot ids with their own region id, as a hole fetched on its own does. They shared one unprefixed sequence, so a slotted island in a batched hole could take the page's own slot id and adopt the wrong children.
  • #1695 3faef3d Thanks @PuruVJ! - In dev and rules builds, a batch of holes where one hole skips its page's data and another asks for it no longer breaks rule_hole_page_read_unpredicted when the page's data could not be looked up.
  • #679 79f67f7 Thanks @PuruVJ! - A batch's answer for a hole that lands after a newer refresh of it no longer paints its older HTML over the refresh.
  • #918 b9af6be Thanks @PuruVJ! - Holes that start together are batched only with region calls of their own endpoint path; a plugin's hole (a deferred peer widget) goes on its own instead of being posted to the wrong endpoint and refused.
  • #717 bf92b58 Thanks @PuruVJ! - A batch of holes is charged to the visitor's render budget once per hole it renders, after that hole's capability verifies, the same as each hole's own request; one call asked twice in a batch renders once. A hole over budget is left out of the batch and fetches on its own.
  • #1451 032d21f Thanks @PuruVJ! - A request body that breaks off mid-upload is a 400 on the profiler's login and on the federation routes (the exposed table, a thaw notice), not a thrown error and a 500.
  • #442 95ed95c Thanks @PuruVJ! - say() in @ogygia/machine says a thing once per key until it recovers, remembering at most 1024 keys; a machine's warn-once and a plugin's warn use it, so a long-lived server no longer keeps every distinct warning text.
  • #1114 36de4b8 Thanks @PuruVJ! - A region capability's exp and ttl are accepted only as ogygia mints them (plain digits, no leading zero) and refused as shape before the signature check otherwise (before, any spelling of the same expiry, such as +n, 0n, n.0 or 0x…, verified, so one capability had many URLs).
  • #926 9caa92e Thanks @PuruVJ! - A server island's address is a path from the app's root (/__ogygia__?…, under paths.base when set), never relative to the page, so it still works in HTML that moves to another page or app under Kit's default paths.relative.
  • #1095 b3f4d94 Thanks @PuruVJ! - The region batch endpoint and federation's routes stop reading a request body once it passes their size cap, also when it is sent without a content-length (before, such a body was read whole, then refused).
  • #1176 25db3c9 Thanks @PuruVJ! - A plugin whose browser boot throws no longer stops the runtime's boot: it is reported once (client_hook_failed, hook boot), the other plugins still boot, and the page's regions still wake.
  • #730 f7ffd0e Thanks @PuruVJ! - A build now fails with secret_in_client_chunk when a client chunk or asset holds a signing secret (OGYGIA_SECRET, or the build's own key) or a server module that signs region addresses.
  • #1368 574999e Thanks @PuruVJ! - A runtime script the page emits loses every code-origins attribute it carried, however it is written (single quotes, no quotes, spaces around =, twice), so the origins the render learned are the ones the browser reads.
  • #986 d460c55 Thanks @PuruVJ! - @ogygia/core: in dev, a region's style module is imported only from the page's code origins, and a dropped download is asked for again only there, the same check an island's own code load uses (#981).
  • #985 9459fff Thanks @PuruVJ! - A stylesheet an island's code put in the head as it loaded (the bundler adds the CSS the island's chunks need, as on a mounted peer page) leaves with the page after a router navigation, as a fresh load of the next page has none, and comes back with a page whose island runs that code again.
  • #1403 891228b Thanks @PuruVJ! - A hydration mismatch report drops a comment where the browser ends it (--!>, <!-->), and keeps a <!-- inside a script or a quoted value.
  • #1278 925cf79 Thanks @PuruVJ! - componentCssHead no longer links every sheet again when a request asks for the same components a third time.
  • #1541 1a475de Thanks @PuruVJ! - A context that cannot cross into regions is now said, not dropped silently: a one-off Symbol() key (context_key_one_off) or a function value (context_value_function) set with setContext from ogygia or in a <Provide> warns in dev, naming the key and what to use instead; the build warns on a one-off Symbol() key it can see (region_context_key_one_off), and its "island code reads a context nothing bridges" warning now covers Symbol.for(name) keys.
  • #1541 1a475de Thanks @PuruVJ! - A Symbol.for(name) context key, set with setContext from ogygia or in a <Provide>'s values, now reaches regions in the browser: it crosses by its name. Before, the drop-in setContext left it out and a <Provide> holding one failed to render.
  • #474 e891f09 Thanks @PuruVJ! - Render deadlines, the region endpoint's rate-limit windows, a lake's maxAge and the page-asset budget are now measured on the monotonic clock, so a wall-clock step (an NTP correction, a VM resume) no longer shortens or stretches them.
  • #1887 2fdd64b Thanks @PuruVJ! - Region's generated .svelte.d.ts no longer names RegionValue, Snippet and Component without importing them, so it type-checks with skipLibCheck off; its props are unchanged.
  • #1483 25c95f5 Thanks @PuruVJ! - A hole answer redirected to another origin is refused like a same-origin redirect: no retries, and dev says region_answer_redirected with the URL that answered.
  • #852 a3ca2ad Thanks @PuruVJ! - A client navigation to a page with more copies of an island than the current page shows live now gets the extra copies rendered by the server, instead of empty placeholders the browser had to fill.
  • #1618 80f4d91 Thanks @PuruVJ! - With serverDelta on, a page the router kept (a soft refresh's, a prefetch's) is reused only by a document that still shows the regions its fetch left out: going back to a page after a soft refresh on it no longer brings its regions in empty and fails to hydrate them.
  • #1460 3ddf36b Thanks @PuruVJ! - In dev, a route group whose name holds a quote mark or an ampersand gets its edited stylesheets live again: the dev-scope meta now escapes the name instead of dropping its quotes.
  • #1966 c22268b Thanks @PuruVJ! - A build that ships devtools events on every page (a plugin's events() without lazy) now marks each page cache-control: private, no-store, so a shared cache can't keep one visitor's server events and serve them to another.
  • #882 724b0f3 Thanks @PuruVJ! - Devtools builds: the devtools stream takes server events only from the response's own side channel, not from one the page's markup carries.
  • #880 5373f25 Thanks @PuruVJ! - On a devtools build, only well-formed server events from the page's devtools side channel enter the event stream, and ingest() (a saved trace, another realm) takes only well-formed events of either realm; anything else is left out.
  • #1555 65733b6 Thanks @PuruVJ! - The check for edits to a sleeping island now watches only the islands still asleep, not the whole document, so DOM changes elsewhere on a large page (a framework hydrating it, web components rendering) no longer run the runtime's code.
  • #594 0a40057 Thanks @PuruVJ! - @ogygia/core: a click made while an island is starting goes to the island alone; the page's own click listeners hear it once, as the replay when the island is live (late, and with isTrusted false), not at the press and then again at the replay.
  • #505 abc4a42 Thanks @PuruVJ! - In dev, a file added next to a barrel that already re-exported it is now picked up by the barrels pass and the CSS walk, without restarting the dev server.
  • #1190 1bc6577 Thanks @PuruVJ! - A whole-body swap navigation keeps the objects the next page's seed revived: an island that decodes the same object as page.data gets that object, not a fresh copy.
  • #1348 e0b4429 Thanks @PuruVJ! - An og.$ factory whose code holds </script or <!-- no longer breaks the page in a build: the script that registers the factories keeps it inside its tag.
  • #705 79e5bfa Thanks @PuruVJ! - An import.meta.og.$ function now travels as its build tag and captures only, never its source: a side whose build lacks the tag refuses it with fn_revive_missing instead of rebuilding it from the payload.
  • #2023 00a7bc4 Thanks @PuruVJ! - A peer's live region or kept island, mounted in a shell, is hydrated by the peer's own build, and its next tick or page's props reach it in place (a peer on an older build is hydrated again instead).
  • #1970 f9ebb02 Thanks @PuruVJ! - An island from a mounted peer page now reads the cross-island context its page provides (<Provide>, setContext, createContext), even when the shell provides no context of its own. Before, a production shell left the context bridge out and never handed context to a peer's island, so the island hydrated with its context silently empty. A federated build now also ships the live-region feature a peer's held or kept regions need.
  • #1338 0a4a20e Thanks @PuruVJ! - A form value kept across a navigation goes back into the island it was typed in, found by its component and which copy of it it is, even when the page now has other islands in front of it.
  • #1172 c9e4346 Thanks @PuruVJ! - A router link whose hash is a bad percent escape (#%E0) no longer breaks the navigation after the swap: the router finds the hash target as the browser does (the hash as written first, then decoded, never throwing), so after hooks and og:page-load run and the router goes idle.
  • #916 daa8d5c Thanks @PuruVJ! - A hole's frame address now names the whole call it asks (the endpoint path and every parameter but the capability's own), so two plugin holes alike but for their props, like two deferred peer widgets, no longer share one answer, on one page or across a navigation.
  • #1029 40b91ab Thanks @PuruVJ! - @ogygia/machine: A cache() set of the value its key already holds keeps it (no 'delete' notice, so an onEvict cleanup no longer runs on a value still held), set takes its own keep, and a keyed flights of one field keys by that field's value with no string built. @ogygia/core: A region answer written to the frame store costs about a third fewer steps (one cache write, no lookup before it), and an island warm reads its chunk graph once.
  • #1353 fc49f63 Thanks @PuruVJ! - goto() returns a rejected promise for a URL it refuses (not http, another origin without external, unparsable) instead of throwing, as Kit's goto does: goto(x).catch(...) now sees it. A try { goto(x) } catch without await no longer catches it.
  • #1002 13e08a2 Thanks @PuruVJ! - @ogygia/core: an island's graph preload no longer adds a second <link rel="modulepreload"> for a chunk that another loader (such as Vite's preload helper) linked after the first island started. It now checks the document's links each time it preloads.
  • #1457 620d27e Thanks @PuruVJ! - The head's end is found as the browser reads it: an app.html that writes </HEAD> gets ogygia's head injections, and a </head> inside a head script's text (a JSON-LD string) no longer takes them into the script. The body's end is found in any case.
  • #825 6f03b81 Thanks @PuruVJ! - A client navigation to a page at another URL depth no longer adds a second copy of a head tag (the ogygia runtime script, a stylesheet) that the first page wrote with a page-relative URL.
  • #576 7be0138 Thanks @PuruVJ! - The head's duplicate-link pass reads each <link> attribute as HTML does, so a stylesheet or preload hint whose URL has an apostrophe is no longer dropped, and data-rel / data-href no longer count as rel / href.
  • #997 f018ab4 Thanks @PuruVJ! - @ogygia/core: the server's head checks, link dedupe and runtime placement, and the router's prefetch warm read a page the way the browser does: a tag written inside a comment or inside a script's, title's or textarea's text no longer counts, so the runtime is never moved into a commented-out script or a title, a real stylesheet is never dropped as a copy of a commented one, and text that looks like a region no longer makes the router import a module.
  • #826 715744f Thanks @PuruVJ! - A client navigation that keeps a region (a refresh in place, or a link to a page with the same region) now keeps the head tags that region's inner islands wrote, such as the ogygia-kit-island stamp and their stylesheets; each such tag names its owning region in a data-og-owner attribute.
  • #946 b54c1bd Thanks @PuruVJ! - A held region awaited inside another held region's island no longer stays empty when the island re-creates it in the browser: the fresh region paints its value.
  • #1211 c52057f Thanks @PuruVJ! - A held region rendered in a load or by a remote call during the page render no longer mints slot ids the page's own regions also use, and no longer skips more copies on a server-delta navigation than the browser has live.
  • #905 07c4c5d Thanks @PuruVJ! - A server island (render: 'deferred') inside a held region that wakes now shows: the region's HTML is rendered the way its component hydrates, so the server island renders inline as it does inside any island, instead of an empty element that was never fetched.
  • #1252 527fb37 Thanks @PuruVJ! - A page that awaits a held region with an island inside ships the runtime script again with the router off: the island in the held region's own render no longer takes the page's one runtime claim.
  • #1089 c4a1b4a Thanks @PuruVJ! - A router navigation from a styled held region to an unstyled one in the same place no longer keeps the first region's <link data-ogygia-region-css> inside the second: a region's sheet counts as render output in the morph.
  • #440 4a038b4 Thanks @PuruVJ! - A cached hole (maxAge) that reads a flag, or a cookie, header or locals value, for an anonymous visitor is no longer kept and served to other visitors.
  • #830 d5a120a Thanks @PuruVJ! - On a Kit-hydrated page, a deferred region whose props hold a shared object (a wired class instance, a store) now has the same identity in the browser as on the server, so a region Kit mounts afresh finds its server-minted address and loads.
  • #952 7e9f615 Thanks @PuruVJ! - A server island inside a server island that sits in an island no longer waits for that island to wake: it fetches and shows on its own schedule, even when the island waits for a press.
  • #1767 f7e0684 Thanks @PuruVJ! - @ogygia/core: a deferred region refused with sig, shape or foreign-key (a server of another build or key, as in a rolling deploy from 0.12.x) asks once more after half a second, so a load balancer can route it to a server of the page's own build; if that is refused too, region_answer_denied says so and the fallback stands.
  • #1032 4c732c6 Thanks @PuruVJ! - A hole with maxAge whose render reads page.data or calls a remote query is kept in the server's render cache again: the page it renders for is part of its key, not a read about the visitor.
  • #680 f8a4fbf Thanks @PuruVJ! - A hole that a region refresh gives new props while its first fetch is still out now fetches and shows the new answer, instead of keeping the old props' answer.
  • #967 8ef4954 Thanks @PuruVJ! - A deferred region the endpoint rate-limited no longer gives up for good: the endpoint's 429 now carries Retry-After (the time left in the visitor's window), and the region asks again after it (at most a minute) instead of spending its tries inside the window.
  • #1210 a365b26 Thanks @PuruVJ! - A hole whose server render throws is now said in production too (region_render_failed, with the error, in the server's log). It was said only in dev, so a hole that failed only in production answered 500 with nothing logged. The visitor's answer is unchanged: an opaque 500.
  • #897 226364d Thanks @PuruVJ! - A deferred region whose fetch fails while the browser is offline, or an island whose code fails to download then, now tries again once the browser is back online instead of giving up for the life of the page; a region that finally gives up says region_fetch_failed in production too.
  • #446 37a9eaf Thanks @PuruVJ! - A hole whose prefetch warm still waits for a background turn fetches at once when it wakes on screen, the visitor reaches for it, or preload() asks for it, instead of waiting behind the page's other prefetches.
  • #855 db021f1 Thanks @PuruVJ! - @ogygia/core: island code loads only from the page's origin, the runtime's own, and the origins the server names on the runtime's script (a federation peer's); an entry or src anywhere else is refused with island_code_refused, and a region's src re-points only its own island's built file (#851).
  • #1203 2378a3c Thanks @PuruVJ! - A production server that cannot find or read the client build's handoff (og-region-deps.json) now says so (island_deps_handoff_missing, with the paths it tried) instead of rendering every hole without its CSS and preloads in silence. A handoff that failed to read or parse is tried again on the next request instead of being replaced by an empty one for the life of the process.
  • #838 b823bff Thanks @PuruVJ! - @ogygia/core: the island graph keeps every entry whatever its name; one named __proto__ was lost on the way to the page (#994).
  • #1119 4c64ac7 Thanks @PuruVJ! - On a csr=false page, an island's <svelte:head> content stands in <head> where the server wrote it, whatever order the islands wake in, so two islands' head styles that set one property no longer swap winners from one load to the next.
  • #1125 17834e0 Thanks @PuruVJ! - A <style> written over several lines in an island's <svelte:head> is one copy once the island wakes: the page's server copy is retired, as it is for single-line head content.
  • #1924 7929876 Thanks @PuruVJ! - An island whose code failed to load during a short outage wakes once the connection is back. It used to look up the build map first, and if the connection came back during that lookup, the island ended as failed until the page reloaded.
  • #1309 3fde46a Thanks @PuruVJ! - The islands endpoint is decided once per server, by the first handle() (or the first render when none named one), and never moves: a later handle() naming another one, from a second copy of ogygia too, is islands_endpoint_conflict (thrown in dev, said in production) and the first path stays, so the pages already minted and the handle that answers them never name two paths and holes no longer come back empty.
  • #1197 115137c Thanks @PuruVJ! - The islands endpoint answers on its one path under the app's base, the path every capability is minted for: with handle({ endpoint: '/api' }), a page at /docs/api is a page again, not taken for a hole render. A second handle() naming another endpoint in one production server is now said (islands_endpoint_conflict) instead of silently replacing the first.
  • #1086 9f87204 Thanks @PuruVJ! - A head link the router keeps across a client navigation to a deeper or shallower path still names its file: the router reads each page-relative head URL (Kit's default ../_app/...) against the address its page wrote it for and leaves loaded links as they are; only a parked (disabled) sheet, not fetched yet, is made root-relative before the address moves.
  • #1088 59495f6 Thanks @PuruVJ! - A region kept across a client navigation (keep: 'name') now carries the next page's data-og-fp along with its props, so its props sidecar is found and the router lists the islands the page really holds.
  • #841 99e9496 Thanks @PuruVJ! - An island waiting for SvelteKit on a csr=true page no longer waits forever when Kit's own client files fail to download: it fails, and the page says kit_boot_failed once.
  • #2015 19f3a73 Thanks @PuruVJ! - An island waiting for Kit's page on a Kit document now ends its wait with kit_boot_failed when one of the page's own Kit files (a route node) fails to load, not only Kit's entry files: it no longer waits forever.
  • #841 99e9496 Thanks @PuruVJ! - An island waiting for SvelteKit no longer imports a file named by a boot-shaped inline script that came in as markup: only files on the page's or the runtime's origin are imported.
  • #1748 74a5864 Thanks @PuruVJ! - On an app with no router, a page that shows a mounted peer page's islands (a mount.kit catchall) loads the runtime, so those islands wake: no runtime was shipped when the page rendered no region of its own.
  • #1747 bf5db5d Thanks @PuruVJ! - A Kit page's runtime script names the code origins its render learned (a mounted peer page whose runtime is served from another origin): they were dropped, because Kit runs the page transform outside its request store.
  • #1856 a878cae Thanks @PuruVJ! - In dev, a csr=true page whose SvelteKit start script ogygia cannot hold until its runtime booted now warns (kit_start_ungated) instead of letting Kit start first without a word.
  • #1711 a44b2c4 Thanks @PuruVJ! - On a csr=true page with regions, Kit now starts only after ogygia's runtime is ready, so lakes keep their content when Kit remounts the page; before, a fast Kit start could leave every lake empty. Kit still starts if the runtime fails to load, and a hash-based CSP gets the hash of Kit's changed start script.
  • #892 4d7af4c Thanks @PuruVJ! - The server takes an SPA nav's live-copy header over the size the router ever sends as no claim, and renders every region.
  • #1566 c7c1514 Thanks @PuruVJ! - Two lakes of the same component with different props each keep their own content when Kit remounts a csr=true page; before, the second showed the first one's HTML and lost its own islands.
  • #1229 e2ea18d Thanks @PuruVJ! - In dev, an island whose hydrate renders a different number of lakes than its server HTML had says so (rule_lake_restored) instead of dropping a lake's content silently.
  • #1443 94f89d5 Thanks @PuruVJ! - A lake whose host island puts its content back is settled at once when a region inside it waited: a later refresh of that lake no longer re-enters its regions as if they still waited.
  • #1692 a8c1cb4 Thanks @PuruVJ! - An island's hydrate no longer moves its lakes out and back: a custom element in a lake stays connected through the hydrate, and what its own runtime did to its light DOM is no longer taken for drift.
  • #1364 33dc452 Thanks @PuruVJ! - A streamed page's late chunk swaps into its slot when one of its sheet hrefs or its slot id holds a quote or a backslash; before, it stayed out, or its sheet was added twice.
  • #1889 81e1725 Thanks @PuruVJ! - An island a client navigation loads late no longer reports island_load_failed when its load fails after the island was already taken off the page.
  • #594 0a40057 Thanks @PuruVJ! - A lazy chunk whose download failed is asked for again in Safari too, and in Firefox a chunk whose dependency failed is no longer retried as that dependency (which ran a second copy of it and broke the wake): the build hands each lazy loader its own file.
  • #595 98730f8 Thanks @PuruVJ! - A lazy chunk's failed retry fails its caller at once, and a use that comes before the runtime knows why it failed waits for that answer instead of failing with it (in Safari, a click right after a failed prefetch now navigates in place).
  • #2039 e55454d Thanks @PuruVJ! - A remote query refreshed by an island's own handler after the island left the page (a click awaiting a command, then a navigation) is a no-op instead of Kit's "No cached query found" error; the next page still reads that query fresh.
  • #782 bab2aff Thanks @PuruVJ! - A static live region put back on the page after a minute away keeps the newest answer it showed, instead of repainting its last tick over it.
  • #823 6bd663c Thanks @PuruVJ! - A live query in an island now keeps updating after a client navigation that keeps the island; the navigation releases only the query instances that departed islands held.
  • #857 c19ba92 Thanks @PuruVJ! - A live query that a kept island reads only from a component it mounts later keeps running after a client navigation in which another island that read it leaves.
  • #943 08e5c16 Thanks @PuruVJ! - A live region whose island file cannot load stops trying on every tick: it says island_load_failed once, shows each later tick's server HTML, and no longer throws the import's error at the page or asks the server about the file again.
  • #683 0e4f3a8 Thanks @PuruVJ! - A static live region no longer paints a tick over a newer update of the same region (a command that returned it) that landed while the tick was on its way, nor a tick whose data version is older than the one on screen.
  • #491 1e96c8c Thanks @PuruVJ! - Two builds of the same code write the same og-locations map under the same name, so a deploy with no island change no longer busts its cache.
  • #878 08fd293 Thanks @PuruVJ! - A hole answer's region URLs, island graph and region CSS are read as a browser reads the tags: a region or sheet inside an HTML comment is left alone, and upper-case tags, single-quoted values and a quoted > no longer hide one.
  • #1116 9f7eac5 Thanks @PuruVJ! - @ogygia/core: Reading tags out of server HTML (the head dedupe, region CSS, hole answers, seed injection, stitching) costs 2 to 53% fewer steps: each tag is read once, a character at a time. @ogygia/federation: Lifting a peer page and pinning its asset refs costs about half the steps: one walk, each tag's attributes read once.
  • #999 97d59ed Thanks @PuruVJ! - @ogygia/core: holes that read page.data show the data of the page a router navigation leads to, not of the page it left. The navigation's hole batch now names its destination as its referrer.
  • #999 97d59ed Thanks @PuruVJ! - @ogygia/core: a router navigation groups the next page's deferred holes by endpoint, as the first load does. A peer widget's hole no longer sends the whole batch to its own path, which answered 403, and no region call is sent to an endpoint that does not know it.
  • #1048 f848e22 Thanks @PuruVJ! - A page reached by a client navigation loads island code from the code origins its own response names (a federation peer's CDN), instead of refusing them until a full reload.
  • #1963 2bd2d0a Thanks @PuruVJ! - The devtools now say opt-out when a navigation swaps the whole page because the page opted out of the in-place update with data-og-no-reconcile. It used to say shadow-region.
  • #797 69c3a26 Thanks @PuruVJ! - After a client navigation, the next page's head nodes (its <style> and stylesheet <link> among them) sit in the order a fresh load has them, after the app's global styles, so the cascade is the same however the page was reached.
  • #785 1e21a17 Thanks @PuruVJ! - A client navigation to a page with no <title> no longer leaves an empty <title> in the head; the page matches a fresh load of it.
  • #1453 16b6c56 Thanks @PuruVJ! - An island nested beside an {@html} that a third-party script edits before the islands wake is put back from the server markup at the wake, instead of the outer island throwing its hydration away.
  • #1522 9299656 Thanks @PuruVJ! - Dev no longer warns that a nested island's schedule is ignored when it is the same schedule as the island around it (a wake: 'load' island inside a wake: 'load' island, such as one rendered in a mobile menu sheet).
  • #1214 ecd3b32 Thanks @PuruVJ! - data-og-no-reconcile on <body> itself now opts the page out of the router's in-place reconcile, as it does on any element inside the body.
  • #485 ee4cebc Thanks @PuruVJ! - import.meta.og.bake addresses each evaluated bundle by a 64-bit hash of its inputs (was 32-bit), so two modules can no longer share a cached bake by collision.
  • #719 f878494 Thanks @PuruVJ! - ogygia.files checks that each declared path stays inside its package by the path the file really has, so a symlink in the package that leads out of it is refused (package_files_escapes) or left out of a glob.
  • #577 13aa062 Thanks @PuruVJ! - A $page.data promise nested more than 65 levels deep, or inside a cyclic value, now streams into regions like any other: before, it was left in the seed, the page failed to serialize it, and a rejection went unhandled.
  • #577 13aa062 Thanks @PuruVJ! - On a router (non-navigate) request, $page.data with promises keeps its key order in the seed; before, keys came in the order their values settled.
  • #1436 8501aa3 Thanks @PuruVJ! - A streamed page.data promise read again after its value arrived (two regions sharing it, a lazily woken island) is the same settled promise, no longer rejected as page_promise_unresolved.
  • #1627 b43988f Thanks @PuruVJ! - A hole or remote call rendering for a page whose data answer breaks before its first line now renders without the page's facts, as for any other failed look-up, instead of failing.
  • #1000 a29b558 Thanks @PuruVJ! - @ogygia/core: a hole or remote-rendered region that reads page.data always renders with the facts its page look-up returned. Before, it could silently render with an empty page.data when the look-up's answer was not kept.
  • #1768 45632dd Thanks @PuruVJ! - @ogygia/core: the page.data keys a deferred region's answer brings are no longer lost when an island that woke before the answer hydrates first: they wait for devalue to load, then join the page, so islands inside the region read them.
  • #1190 1bc6577 Thanks @PuruVJ! - A region that reunites with an object another region decoded now owns it too, so a navigation that removes the first region no longer releases the object the kept one still shows.
  • #1190 1bc6577 Thanks @PuruVJ! - A client navigation releases the page objects nothing on the next page holds (a seed's values, a decode outside any region), so the hub no longer grows with every page visited in the tab.
  • #1302 0845054 Thanks @PuruVJ! - Dev now says page_payload_unreadable, with the error, when the browser cannot read back data the server wrote for islands (the page seed, page facts, the holes record, the context marker, remote functions' data). Islands used to go on without it and nothing was said; the usual cause is a custom type the server's transport encodes with no decoder in the browser.
  • #1327 239e0e7 Thanks @PuruVJ! - A streamed load promise that never settles no longer holds the server's response open after the visitor leaves: the wait ends with the request and the response closes. A region rendered outside its page now gets a rejected promise (page_promise_unresolved) instead of one that stays pending when Kit's data answer ends without its value. Dev warns (page_promise_pending) when a streamed promise is still pending after 10 s.
  • #1318 295b9a8 Thanks @PuruVJ! - A streamed page.data promise in an island no longer waits forever when its value can no longer come: once the page has loaded with no value for it, or its value came but the code to read it never downloaded, it is rejected with page_promise_unresolved and {#await} shows its catch branch.
  • #1133 8686f2e Thanks @PuruVJ! - A region that leaves the page while its answer waits on its stylesheet no longer stays "painting" for good: its paint ends with the region. In dev and in OGYGIA_RULES=1 builds, a new rule rule_region_left_busy reports a region that left the page with one of its machines still mid-work.
  • #838 b823bff Thanks @PuruVJ! - @ogygia/core: the batch stream reader skips any number of malformed parcel openings in a loop, instead of running out of stack.
  • #912 4cfe3ea Thanks @PuruVJ! - A federated app's production runtime now ships what a peer's regions need (the frame store, morph, lakes, wakes), so a peer's render: 'deferred' widget loads in an app with no deferred region of its own.
  • #958 9e1480b Thanks @PuruVJ! - A peer's island (fragment federation) whose code download fails once is asked for again under a fresh URL, like the app's own islands, instead of never loading. A retry goes only to the origin the island's file was first asked from.
  • #1012 26f7e95 Thanks @PuruVJ! - A deferred peer region (and any plugin capability) names the key that signed it, as a region hole does: a server of another deployment (a rolling deploy without one shared OGYGIA_SECRET) refuses it as foreign-key, with the advice that goes with it, instead of as a bad signature.
  • #1819 dcecc57 Thanks @PuruVJ! - A plugin's handle, and a transformPageChunk passed to resolve() or a plugin's next(), are called on their own object, so one written as a method can use this.
  • #1470 a13ad32 Thanks @PuruVJ! - A plugin's signed hole (federation's /og/frag) now pays the visitor's region rate budgets as a region hole does: the probe budget before the signature check, the render budget after. A federation hole can now answer 429 with Retry-After under load.
  • #1359 5ff44f8 Thanks @PuruVJ! - In dev, a plugin's signed hole (a deferred federation widget) no longer answers 429 after the server hooks module runs a second time.
  • #802 19fa62d Thanks @PuruVJ! - core: a plugin's app.sign, app.verify and app.capability sign in the plugin's own domain (its name, length-prefixed), so nothing a plugin signs checks as a region link or as another plugin's, dotted names included; plugin capability links minted before the upgrade are refused once and fall back like any refused hole
  • #747 771fa53 Thanks @PuruVJ! - @ogygia/postrender: Breaking: a page rendered for a request that carries a cookie or an authorization header is not stored (it is still served a stored copy, and its own render is stamped private, no-store); a route whose pages are the same with them opts in with config = { postrender: 'with-credentials' }. @ogygia/core: a render's per-visitor reads also count a function in locals, in and the keys of locals, the request headers read whole or copied while one is personal, headers.has() and a request clone(), client address, forwarded host, location and session headers, getClientAddress(), per-request platform data, and an own-origin fetch that carries the visitor's cookies.
  • #443 8b120e2 Thanks @PuruVJ! - An island's code that a page prefetch started downloading at low priority is raised when the island wakes on screen or the visitor reaches for it, and a hover no longer waits behind the prefetch it repeats.
  • #712 5320d28 Thanks @PuruVJ! - preload(region) sends no request when the region's address is another origin; every fetch of a hole now checks its endpoint is this origin before it goes out.
  • #760 90a7449 Thanks @PuruVJ! - devtools: in production the profiler accepts only a secret of at least 16 bytes; a shorter one keeps its pages closed and says so (devtools_profiler_secret_short) devtools: the profiler allows ten wrong keys per address in 15 minutes (the login form, x-profiler-key and x-profile share the count), then answers 429 until the window passes devtools: a profiler login is a random session id the server signs; it expires after 12 hours, logout ends it, and any change to it is refused (everyone logs in once again after the upgrade) devtools: the profiler's beacon flag cookie belongs to one login: it ends when that session expires or logs out
  • #1964 733aa1c Thanks @PuruVJ! - The profiler now names seed references as the reason an island's props went to devalue when the props themselves were plain JSON. It used to show no reason at all.
  • #1507 9762036 Thanks @PuruVJ! - A hole whose island module throws as it loads, or whose props do not decode, is said in the server's log in production too (endpoint_props_decode_failed), not only in dev. The visitor's 403 stays opaque.
  • #594 0a40057 Thanks @PuruVJ! - A click on an island that goes live while its wake is still ending acts once: the click queued for the replay no longer also reaches the island's handlers.
  • #1301 cb0633e Thanks @PuruVJ! - Past the region endpoint's rate-limit cap, an address whose window already ended is let go first: a visitor who spent their budget in a window still open no longer gets a fresh one early.
  • #1269 94f10fe Thanks @PuruVJ! - A visitor whose rate-limit window rolled over counts as recently seen, so under pressure it is no longer the first address let go (and handed a fresh budget).
  • #980 6856e83 Thanks @PuruVJ! - When the region render budget runs out for a private, loopback or link-local client address (a proxy's, shared by every visitor behind it), the server warns once with rate_limit_shared_address.
  • #529 32177c7 Thanks @PuruVJ! - - @ogygia/core: four new runtime hooks for a plugin's browser half: regionWoke, regionMismatched, navStarted and loadStarted; navigated now says whether the regions were reconciled, and holeFailed carries the hole's el. - @ogygia/devtools: the dock's Page, Hydration and Session readers and the testing API read the runtime hooks instead of core's event layer. - @ogygia/insights: a recorded session holds the runtime hook calls (SessionData.hooks, HookEntry) instead of core's events.
  • #829 0767267 Thanks @PuruVJ! - The read set (facts.reads) is a plain shape, so core's built types and a plugin's agree on it.
  • #709 62e47da Thanks @PuruVJ! - A batch of holes is sent only to this origin's endpoint and writes nothing from an answer that ended on another origin, and invalidateAll() merges a refreshed page only when its fetch landed on this origin. Every string the runtime turns into DOM now goes through one module.
  • #714 9c93fb1 Thanks @PuruVJ! - Every region response now says how it may be cached: a keepFallback() answer (204) carries its hole's own policy (no-store, or private, max-age when the hole opted in), and refusals and errors (403, 405, 400, 413, 429, 500) are no-store.
  • #778 9b9f253 Thanks @PuruVJ! - core: region links are signed through capability(), naming the islands endpoint they are for and refusing a lifetime longer than the app ever mints; links stay identical between renders, and links signed before (pages cached before the upgrade) still work and are re-signed in the new form when renewed machine: capability().mintUntil() mints a token that expires at a given time, so tokens without a nonce stay identical across a window the caller picks
  • #1371 1073b6a Thanks @PuruVJ! - A region import mark that would do nothing now fails the build: render: 'static' alone, keep on a hole, lake or live region, and a call-shaped schedule like wake: 'visible(200px)' (a media query is now media words around balanced conditions, everywhere a schedule is read).
  • #780 71ff4c7 Thanks @PuruVJ! - core: a region request (a hole's GET, its renewal, each hole of a batch) walks one state machine, so rendering or charging the render budget before its capability verifies cannot happen
  • #800 a14a348 Thanks @PuruVJ! - A deferred region's stylesheet leaves <head> with the page that used it, instead of one navigation later.
  • #1852 db542a7 Thanks @PuruVJ! - In dev, region.snippet(snippet) now freezes a snippet that writes elements; before, Svelte's dev element check found no component around it and the freeze threw.
  • #2040 411a3eb Thanks @PuruVJ! - A hole given a snippet prop now works in the browser, as it does on the server: created fresh in the browser it freezes the snippet to HTML itself, and hydrated on a csr page it keeps the server's frozen HTML, instead of throwing prop_not_serializable. A top-level island given a snippet prop and created fresh in the browser now freezes the snippet to HTML, as the server does, instead of throwing prop_not_serializable.
  • #2017 f54b0d7 Thanks @PuruVJ! - A region's relative entry, src and endpoint are pinned against the address its document was loaded at: a back taken before the runtime ran no longer points its island code one folder too deep.
  • #1199 fff7b91 Thanks @PuruVJ! - A region whose wake is not a trigger now says so (region_wake_unknown, with the region and the value) as it falls back to waking on load.
  • #608 67127a4 Thanks @PuruVJ! - - @ogygia/core: the runtime no longer carries the restore support for an app's own HTML pass (the restore report, a web component's class tokens through hydration, the answer restore hooks); apps without @ogygia/rewrite boot about 0.15 to 0.2 kB (brotli) smaller. - @ogygia/rewrite: rewrite() ships that support to the browser itself; restore_markup_broken and restore_late are its message codes.
  • #757 a403024 Thanks @PuruVJ! - core: the restorer never puts back an event-handler attribute, a srcdoc or a javascript: URL when it returns a host's attributes to Svelte's
  • #834 725eb10 Thanks @PuruVJ! - When the file of an island nested in another island fails to download once, the retry now asks again for the outer island's own file, so the outer island is no longer replaced by the nested one.
  • #1902 84489df Thanks @PuruVJ! - A page with declarative shadow DOM now works after a router navigation: its web components get their shadow roots, no <template shadowrootmode> is left inert, and its islands hydrate instead of being discarded. A host the navigation keeps (same element at the same place on both pages) keeps its live shadow root; give it a different id or data-key on each page to take the new page's render.
  • #2026 face75c Thanks @PuruVJ! - A Back the visitor takes after a reload, before the router has started, no longer leaves the old page on screen under the new address: the router loads the page the address names.
  • #2025 7582a45 Thanks @PuruVJ! - A context set with createContext's setter (or ogygia's setContext) in a server-router page now reaches the islands on that page. A context set with ogygia's setContext reaches the page's islands when the server loads ogygia's handle module twice (the Vite dev server can), not only on the first request.
  • #447 9807aa5 Thanks @PuruVJ! - A click on a link whose page prefetch still waits for its turn sends that request at high priority, the visitor's, instead of the prefetch's low one.
  • #1022 dd93a29 Thanks @PuruVJ! - @ogygia/core: a router return to a page within a minute shows its deferred regions from the answers already held, instead of asking the server for them again.
  • #859 437e91b Thanks @PuruVJ! - In dev and in OGYGIA_RULES=1 builds, a new rule rule_region_answer_behind reports a deferred region that shows an older answer than the newest one its address holds.
  • #524 9df2941 Thanks @PuruVJ! - - @ogygia/core: a plugin's browser half subscribes to the runtime's moments with defineClient({ hooks: { … } }) (region mounted, hydrated, failed, hole answered, navigated and the rest); a hook no plugin subscribes to is left out of the runtime, and regionMount / regionUnmount move inside hooks. - @ogygia/devtools: the profiler's beacon is part of devtools' browser half, so an app without devtools no longer carries it in its runtime.
  • #1886 1b756bf Thanks @PuruVJ! - html and attr read a URL attribute's scheme the way a browser's URL parser does: a value a browser follows as a relative URL (a space before the colon, as in javascript :x, or a leading non-ASCII space) is kept, not replaced with about:invalid#unsafe; every value a browser would run as javascript:, vbscript: or data: is still refused.
  • #1898 adef9d0 Thanks @PuruVJ! - An island prop that matches a page seed node only up to -0 versus 0, or an empty array slot versus undefined, now crosses as its own copy, so the island gets back exactly what it was given.
  • #718 9ef66e4 Thanks @PuruVJ! - The server honors x-ogygia-known (the regions a router navigation already shows) only when the app turned on router: { serverDelta: true }, and a page then answers with Vary: x-ogygia-spa, x-ogygia-known, so a shared cache never serves a copy with regions left out to a visitor who asked for the whole page.
  • #1201 ae5d636 Thanks @PuruVJ! - The server's request ports (the document tail, the page seed and late-region recorders, the context recorder, the devtools recorder, the request stats sink) are one per process, so a second copy of core in the server bundle reads what the handle installed instead of nothing.
  • #721 a9072b3 Thanks @PuruVJ! - A region answer's server-timing (queue and render time) goes only to dev, a browser that turned devtools on, or the profiler's logged-in user; the profiler's beacon flag cookie now carries a value signed under the profiler secret, and a bare 1 no longer turns it on
  • #827 5becc79 Thanks @PuruVJ! - A client navigation that removes one of two regions sharing one live object (a wired class instance, a store) no longer disposes that object while the other region stays on the page.
  • #990 e5f7368 Thanks @PuruVJ! - @ogygia/core: a stylesheet that islands under two different outer islands share stays in the head after a client navigation while any of those islands is still on the page. Before, it named only the first island and could go while the second still needed it. The page's tail now lists the other owners (<script type="application/json" data-og-owners>), and data-og-owner can name several islands, space-separated.
  • #979 1398f5f Thanks @PuruVJ! - Islands read the page seed, remote seed, page context and holes record only from the server's own response, not from a script of the same type the page's markup carries.
  • #1141 3132e4c Thanks @PuruVJ! - A region snippet's slot is found by its id's exact text, with no selector built from it: an id with a quote or a bracket no longer throws or matches another slot, and its fallback marker escapes the id.
  • #1126 c94c8e0 Thanks @PuruVJ! - A soft refresh (invalidateAll) while a navigation waits on the next page's stylesheet no longer takes that sheet out of the head: the navigation goes in as soon as the sheet loads, styled, instead of after the full 2 s wait and unstyled.
  • #784 1497db2 Thanks @PuruVJ! - core: a navigation never adopts a meta http-equiv="set-cookie" from a fetched page (beside <base> and refresh / CSP metas), and a dev and e2e rule (rule_head_adopted) says when the head gains any of them after the first document
  • #594 0a40057 Thanks @PuruVJ! - In dev, a tab whose islands fail to load after a dependency re-optimize reloads once, not once per failing island.
  • #694 4c99764 Thanks @PuruVJ! - @ogygia/core: every answer that reaches a region's address, applied or dropped as older, is now a move of a stored-answer machine, so observe(), a test's machine trace and devtools see each one.
  • #964 3b77e25 Thanks @PuruVJ! - ogygia is now built and tested on Svelte 5.57.3. On Svelte 5.56, a held region whose query fails after another one failed throws a TypeError from inside Svelte; upgrade to 5.57 to avoid it.
  • #1270 11e87f7 Thanks @PuruVJ! - The viewport tracker no longer keeps observing a region with no box of its own after it stops following it (for example once a navigation took the region off the page).
  • #1525 03b75ac Thanks @PuruVJ! - The router reads the address bar for the page on screen, so a page script's own history.replaceState (stripping tracking params, or devtools' ?og-devtools=on|off switch) no longer leaves it remembering the old address for a refresh, a fragment link or a back step.
  • #483 721894d Thanks @PuruVJ! - A hole fetched on interaction and kept as a live region revalidates on load after a remount, instead of waiting on a media query named interaction that never matches.
  • #1479 c807794 Thanks @PuruVJ! - In dev and rules builds, the hydration scheduler's count of waiting islands is checked against the islands it holds (rule_turns_counted), so a drift is said instead of leaving islands waiting.
  • #818 08a09ed Thanks @PuruVJ! - Leaving a page with two identical regions (same component, same props) now removes both of them; one of the pair no longer stays on the next page.
  • #1476 e29dc84 Thanks @PuruVJ! - In dev, a plain error thrown in ogygia's own code that nothing caught (a TypeError, say) now breaks rule_failure_uncoded, not only an [ogygia] message with no code.
  • #1505 8216703 Thanks @PuruVJ! - A streamed page.data promise the page never resolved rejects with ogygia's coded error (code: 'page_promise_unresolved'), not a plain copy of its message.
  • #1729 d3fcc32 Thanks @PuruVJ! - Where there is no IntersectionObserver, a viewport watch still throws to its caller, but no longer leaves a broken tracker behind that fails every later watch and throws when the page's trackers end.
  • #448 ba47506 Thanks @PuruVJ! - A region with no box of its own (its content is position: fixed, or it is display: contents) with a visible schedule now wakes, fetches and takes its hydrate turn when what it renders is on screen, as its download's priority already said; it waited for its own empty box before.
  • #495 6be3ece Thanks @PuruVJ! - ogygia's generated modules answer only to ogygia:<realm>/… ids now; the old virtual:ogygia/… ids are gone, and a portable snippet's scoped class names change once with its new id.
  • #493 42129af Thanks @PuruVJ! - A build fails with virtual_wrong_realm when one of ogygia's server modules would reach the browser bundle (or a browser module the server bundle), at import time and again over the finished bundle.
  • #489 b3bf281 Thanks @PuruVJ! - An import of a virtual:ogygia/… module that does not exist now fails with the virtual_unknown message, instead of an unresolved-import error from the bundler.
  • #492 e858ae2 Thanks @PuruVJ! - The build serves one module per realm, ogygia:server, ogygia:boot and ogygia:region, with named exports, and ogygia/types declares them, generated from the module table so each type is read off the code that makes the value.
  • #1644 77b4f8c Thanks @PuruVJ! - A view-transition name that only the next page has comes back on its element once a client navigation's transition settles, so the next navigation between two such pages animates it again.
  • #1180 5b7f197 Thanks @PuruVJ! - A region whose wake is not a trigger (a typo like visable, or a value computed at runtime through region()'s options) now wakes on load, as an unknown fetch schedule already did. It was taken for a media query that never matches, so the region never woke.
  • #1087 5442971 Thanks @PuruVJ! - A router prefetch no longer imports a lake's or a server-only hole's entry as a module: the page asked the server for /<region id> twice and ended in chunk_reload_needed (a hole nested in an island, after a client navigation).
  • #1311 779fa59 Thanks @PuruVJ! - A kept region whose next page's props carry a live value the build did not predict now shows those props after the navigation: the runtime loads the wire codec first, instead of silently keeping the old props.
  • #1311 779fa59 Thanks @PuruVJ! - An island whose props carry a live value the build did not predict (a forwarded children, a snippet passed as a prop, a spread of $props()) now hydrates: the runtime loads the wire codec when that island wakes, instead of failing with "wired prop but no wire feature". Builds that already ship the codec are unchanged.
  • #812 4282ea0 Thanks @PuruVJ! - A region whose props carry a wired class, a store or an og.$ function keeps the same data-og-fp from one server render to the next, so a navigation that keeps it on screen keeps it mounted with its state, and the server-delta render can skip it.
  • #983 f3f8389 Thanks @PuruVJ! - @ogygia/core: a server island inside a server island that wakes (render: 'deferred', wake, or a held region with a schedule) shows after the outer one wakes, instead of going blank. The outer hole's signed address now says it wakes (&w=1), and the endpoint renders its answer the way the browser hydrates it.

Updated dependencies: @ogygia/[email protected]

@ogygia/data-shake

Minor Changes

  • #1665 096de6c Thanks @PuruVJ! - Breaking: shipping only the page.data your islands read is a plugin now, dataShake() from @ogygia/data-shake/vite, listed beside ogygia() (plugins: [ogygia(), dataShake(), sveltekit()]). Without it every island that reads the page gets all of page.data, and every server hole looks the page up.

Updated dependencies: @ogygia/[email protected]

@ogygia/devtools

Minor Changes

  • #606 1fdd1a0 Thanks @PuruVJ! - - @ogygia/devtools: breaking: the dock no longer installs window.__ogygia_testing or loads a testing engine of its own. @ogygia/testing injects its engine into every page; list testing() from @ogygia/testing/vite in the app under test (the dock alone still feeds an injected engine).
  • #760 90a7449 Thanks @PuruVJ! - devtools: in production the profiler accepts only a secret of at least 16 bytes; a shorter one keeps its pages closed and says so (devtools_profiler_secret_short) devtools: the profiler allows ten wrong keys per address in 15 minutes (the login form, x-profiler-key and x-profile share the count), then answers 429 until the window passes devtools: a profiler login is a random session id the server signs; it expires after 12 hours, logout ends it, and any change to it is refused (everyone logs in once again after the upgrade) devtools: the profiler's beacon flag cookie belongs to one login: it ends when that session expires or logs out
  • #1192 b8d8cf0 Thanks @PuruVJ! - The profiler's /page, /reset, /replay/<id> and /logout take a POST only, so a link on another site can no longer change the profiler's state; a GET answers 405 with Allow: POST. The documented call is the same query sent as a POST: POST /__profiler/page?p=/slow/page&runs=6&format=json&key=<secret>. A run page opened from another site waits for a click before it profiles.

Patch Changes

  • #1387 d16b921 Thanks @PuruVJ! - An island whose hydration changed only attributes is read as the browser reads its tags: a > inside a quoted value no longer cuts the tag, values in any quoting with spaces around = and names in any case are compared, and the first of two same-name attributes is the one kept.
  • #509 936654e Thanks @PuruVJ! - - @ogygia/core: the dock's island-barrel finding says why barrels() left the barrel (it runs code of its own, it is a package the plugin does not list, an import of it marks a region, and so on) with the fix for that reason, and names barrels() from @ogygia/barrels when the app does not list it. - @ogygia/devtools: the profiler's island-barrel warning quotes the same reason, recorded by the build, and its fix names barrels() instead of the removed ogygia({ barrels }).
  • #1138 ad28936 Thanks @PuruVJ! - The profiler's page beacon posts only to an address on the page's own origin, read from the head the server wrote: a beacon tag naming another origin, or one outside the head, sends nothing.
  • #1451 032d21f Thanks @PuruVJ! - A request body that breaks off mid-upload is a 400 on the profiler's login and on the federation routes (the exposed table, a thaw notice), not a thrown error and a 500.
  • #1417 f326128 Thanks @PuruVJ! - The profiler's element counts and unsized images are read as the browser makes them: no tag from a quoted value, a comment ending at <!-->, one element for nested templates.
  • #1328 e6f19d2 Thanks @PuruVJ! - The page tab no longer reports a content-named file as not cached when its cache check failed or was refused (a network error, a 405 for HEAD): such a file is left out of the finding.
  • #475 81a40ea Thanks @PuruVJ! - Durations are measured on the monotonic clock: the memory store's TTLs, a federation deploy check's retry wait, the profiler's recording window, deadlines and busy waits, the testing API's wait, and the perf CLI's waits. A wall-clock step no longer moves them.
  • #1403 891228b Thanks @PuruVJ! - A hydration mismatch report drops a comment where the browser ends it (--!>, <!-->), and keeps a <!-- inside a script or a quoted value.
  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #457 1d836ac Thanks @PuruVJ! - The dev dock keeps the profiles of the 64 most recently profiled pages in memory, and the source-line tables of 256 app files, instead of one per page or file a long session ever touched.
  • #1457 620d27e Thanks @PuruVJ! - The head's end is found as the browser reads it: an app.html that writes </HEAD> gets ogygia's head injections, and a </head> inside a head script's text (a JSON-LD string) no longer takes them into the script. The body's end is found in any case.
  • #1523 682f539 Thanks @PuruVJ! - The devtools launcher rests above a page's fixed bottom bar (a phone layout's bottom navigation) instead of covering its buttons.
  • #878 08fd293 Thanks @PuruVJ! - The profiler's byte breakdown, page seed and standalone export read the page as a browser does: tags inside comments are ignored, and quoted >, single quotes and upper-case attributes no longer mislead them.
  • #871 38a5f65 Thanks @PuruVJ! - The profiler stops unpacking an uploaded .ogp past 256 MB and refuses it with devtools_profiler_upload_too_large (413), instead of unpacking any size.
  • #526 f7a51c3 Thanks @PuruVJ! - - @ogygia/devtools: one page beacon measures every profiled or dock-measured page, with or without the runtime: a Kit-hydrated page now reports the same visit an islands page does (long frames, the slowest interaction with its scripts, fonts, images, the CPU of an interaction), and the runtime's chunk no longer carries the page's vitals, long tasks or CPU sampler. - @ogygia/devtools: the dock starts the page beacon itself when the server did not put one in the page (an app without ogygia's handle), so its findings work there too: in dev at once, in a build (devtools switched on) as a small chunk loaded only then. - @ogygia/devtools: when that chunk fails to download, the page says so once (devtools_page_beacon_load_failed) instead of going unmeasured in silence.
  • #1332 7bcf458 Thanks @PuruVJ! - The Page tab no longer says "this page load was not measured" when reading the measurements failed: it says the read failed, with the error's message.
  • #2046 c2890d3 Thanks @PuruVJ! - An app with the profiler on (devtools({ profiler: true })) builds for production again: the profiler's islands no longer import a server module, which failed the browser build with virtual_wrong_realm.
  • #1097 47b5ba7 Thanks @PuruVJ! - The dock's "profile this page" waits out a busy profiler's long Retry-After up to its two-minute deadline (before, a wait past ~24 days failed the profile with machine_bad_time).
  • #1152 6eacf8b Thanks @PuruVJ! - @ogygia/devtools: the profiler reads each request body up to a cap that fits its route (the login, open before sign-in: 16 KB; the beacon: 6 MB; an uploaded .ogp or a posted report: 32 MB) and answers 413 past it without reading the rest.
  • #1383 782344d Thanks @PuruVJ! - The profiler's render diff and DOM diff cut markup at the tags the browser reads: a < or > inside an attribute value, a comment or a script no longer splits a tag, and single-quoted and unquoted attribute values are compared too.
  • #1108 6ddd1f2 Thanks @PuruVJ! - A report POSTed gzipped to the profiler is inflated under the same 256 MB cap as an uploaded .ogp, and refused past it with devtools_profiler_upload_too_large (413) instead of being inflated whole.
  • #868 c4ea513 Thanks @PuruVJ! - The profiler checks its key and session in constant time on every host, with Web Crypto; on a host without node:crypto login works and no plain comparison is used.
  • #1112 6315434 Thanks @PuruVJ! - The profiler login's next is read as a URL and kept only on the app's origin and under the profiler's path, so a profiler mounted at the root (path: '') no longer sends a logged-in user to another site.
  • #1330 56df4b8 Thanks @PuruVJ! - The profiler's browser store closes its database connection when a read or write fails, so a failed one no longer blocks later opens of the store.
  • #773 671c6ec Thanks @PuruVJ! - devtools: a profiler window id is a signed capability: it expires after 10 minutes, is signed under a key derived for window ids alone, and is checked in constant time; dev with no secret signs under a key of its own process (ids from before this release no longer check)
  • #773 671c6ec Thanks @PuruVJ! - A production profiler with no secret takes no window id (x-og-window): it writes no request to the store under one. The fixed dev key is used in dev only.
  • #529 32177c7 Thanks @PuruVJ! - - @ogygia/core: four new runtime hooks for a plugin's browser half: regionWoke, regionMismatched, navStarted and loadStarted; navigated now says whether the regions were reconciled, and holeFailed carries the hole's el. - @ogygia/devtools: the dock's Page, Hydration and Session readers and the testing API read the runtime hooks instead of core's event layer. - @ogygia/insights: a recorded session holds the runtime hook calls (SessionData.hooks, HookEntry) instead of core's events.
  • #1377 68b8253 Thanks @PuruVJ! - The profiler follows a prerendered redirect page however its meta refresh is written (single quotes, no quotes, spaces around =), and no longer takes text that only names one for a redirect.
  • #1707 b20ae04 Thanks @PuruVJ! - - breaking: an app's own pass over the HTML ogygia sends is the new @ogygia/rewrite package. ogygia.handle({ transform }) now throws handle_transform_moved; list rewrite() (from @ogygia/rewrite/vite, no options) beside ogygia(), and put a rewrite() handle before ogygia.handle() in sequence(). ogygia/markup is removed, and scanRegions / liftRegions / restoreRegions leave ogygia/server. - @ogygia/rewrite: rewrite({ match, render }) renders only the outermost custom elements match picks (a tag prefix, a list of tags, or a test), in one render(elements, ctx) call per chunk of the page and per region answer, so a streamed page keeps streaming and its later chunks are rendered too. - @ogygia/rewrite: each element render gets has ogygia's marks numbered from 0, so a cache can key on its bytes; equal elements in one call are rendered once. - @ogygia/rewrite: ctx carries kind, csr, url, event and signal, an AbortSignal that fires when the request is cancelled. - @ogygia/rewrite: a render's sheets go out once per key per response, into the head while it is open and inline before the first element that needs them after. - @ogygia/rewrite: a failed batch is retried one element at a time; an element that still fails, drops ogygia's marks, or returns HTML that does not close what it opens goes out as ogygia rendered it, with a warning in every build (rewrite_element_failed, rewrite_marks_lost), never thrown. - @ogygia/rewrite: an element split across streamed chunks goes out unrendered, said in dev (rewrite_element_split). - @ogygia/rewrite: several rewrites in sequence() run in order on one marking of the HTML; one listed after ogygia.handle() throws rewrite_after_ogygia, and a rewrite without the plugin throws rewrite_plugin_missing. - @ogygia/rewrite: rewrite({ page }) rewrites the whole document (held until its last chunk) or region answer, says rewrite_page_failed when it throws (thrown in dev), and has .html(str) for a plain string. - @ogygia/rewrite: .check(html, { kind, csr }) runs the whole round trip on one piece of markup for a test, listing failures and, with a DOM, every restored element that differs from Svelte's. - @ogygia/rewrite: a page where nothing was restored ships no restorer, and each streamed chunk that restored something ends with one short restore call. - @ogygia/devtools: the profiler reads a page's regions itself and no longer depends on @ogygia/rewrite.
  • #524 9df2941 Thanks @PuruVJ! - - @ogygia/core: a plugin's browser half subscribes to the runtime's moments with defineClient({ hooks: { … } }) (region mounted, hydrated, failed, hole answered, navigated and the rest); a hook no plugin subscribes to is left out of the runtime, and regionMount / regionUnmount move inside hooks. - @ogygia/devtools: the profiler's beacon is part of devtools' browser half, so an app without devtools no longer carries it in its runtime.
  • #721 a9072b3 Thanks @PuruVJ! - A region answer's server-timing (queue and render time) goes only to dev, a browser that turned devtools on, or the profiler's logged-in user; the profiler's beacon flag cookie now carries a value signed under the profiler secret, and a bare 1 no longer turns it on
  • #1148 addb6e5 Thanks @PuruVJ! - A recorded session (the Record tab, @ogygia/testing's record()) goes on across a full page load when the browser keeps the page it left in its back-forward cache, and a page restored from that cache records on.
  • #1464 e731728 Thanks @PuruVJ! - Stopping a devtools recording no longer removes a console.error wrapper the page added while it recorded (an error reporter keeps seeing console errors).
  • #1380 b41d12f Thanks @PuruVJ! - The profiler's standalone report reads a module script's type and src as the browser does (entities decoded, any quoting) and drops every copy of src, so a repeated src no longer makes the inlined runtime load a file instead of its own code.
  • #1167 562960d Thanks @PuruVJ! - The profiler's standalone HTML export sends the profiler secret only with the report page's own fetch, never with the stylesheets and chunks it inlines (which may be on another host, such as a CDN).
  • #531 dc985cf Thanks @PuruVJ! - - @ogygia/testing: withOgygia injects its own copy of the testing engine into each page before it loads (one bundled script, Playwright's addInitScript), so the engine always matches the package's version. page().self.bytes is 0 then (the page downloads none of it), and page().self.phases.engine is the engine's own start. - @ogygia/devtools: the dock's realm lends the injected engine what only the app's build knows (dev or not, the regions' names, the props lookup), and the leftovers watch keeps one record per page that any copy of the readers can read.
  • #605 ec75d61 Thanks @PuruVJ! - - @ogygia/testing: testing() from @ogygia/testing/vite, listed beside ogygia(), feeds the testing engine a test injects, so page(), hydration(), record() and settled() work without devtools in the app. A visitor runs nothing of it. profile() still needs devtools({ profiler }). - @ogygia/testing: the injected engine starts the page beacon itself, so a page without the devtools server is measured too. - @ogygia/devtools: with both the dock and testing() in an app, each runtime hook call is recorded and measured once.
  • #863 622b5c7 Thanks @PuruVJ! - The profiler's beacon keeps each visit inside one post's bounds however many posts of that visit arrive (before, every post naming new islands grew the kept visit).
  • #1153 b121951 Thanks @PuruVJ! - @ogygia/testing's record() keeps one session across a full page load in WebKit (Safari): the testing engine no longer throws on the new page when it picks up the session the old page saved.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/federation

Minor Changes

  • #776 702f88f Thanks @PuruVJ! - @ogygia/federation: Breaking: an app checks the signed calls it receives against its audience, or the host of its ORIGIN environment variable when it sets none, and never against the Host the request names. With neither, every signed call gets a 401 and federate() warns federation_audience_missing; set audience (or ORIGIN) on each app that peers call into. verify_signed_request() without audience follows the same rule.
  • #776 702f88f Thanks @PuruVJ! - federation: a signed call also carries one capability token (x-og-cap): method, path, body hash, claims and audience as length-prefixed fields, a nonce that checks once, signed with Ed25519; a receiver checks a call that has one by it alone, and still accepts a call with only the legacy headers (x-og-ts, x-og-n, x-og-sig) from a peer that has not upgraded yet
  • #516 368b1da Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking: SharedState is no longer exported from ogygia or @ogygia/core. Import it from @ogygia/federation/shared instead; it works the same. - @ogygia/federation: SharedState (state shared across separate builds on one page) is exported from @ogygia/federation/shared.

Patch Changes

  • #1001 2a2f7ad Thanks @PuruVJ! - A peer's asset refs are pinned to its origin only in the attribute values of real tags: text on a peer page that looks like one (a code sample, a comment, script text) reaches the shell as the peer wrote it.
  • #1121 d61549f Thanks @PuruVJ! - @ogygia/core: Tag and attribute names read from server HTML are lower-cased the way a browser does it, ASCII letters only, so data-Äsrc and data-äsrc count as two attributes. @ogygia/federation: A peer tag with two attributes whose names differ only past ASCII (data-Äsrc, data-äsrc) now has each value pinned in its own attribute. Before, the first value was written into the second.
  • #1451 032d21f Thanks @PuruVJ! - A request body that breaks off mid-upload is a 400 on the profiler's login and on the federation routes (the exposed table, a thaw notice), not a thrown error and a 500.
  • #1095 b3f4d94 Thanks @PuruVJ! - The region batch endpoint and federation's routes stop reading a request body once it passes their size cap, also when it is sent without a content-length (before, such a body was read whole, then refused).
  • #475 81a40ea Thanks @PuruVJ! - Durations are measured on the monotonic clock: the memory store's TTLs, a federation deploy check's retry wait, the profiler's recording window, deadlines and busy waits, the testing API's wait, and the perf CLI's waits. A wall-clock step no longer moves them.
  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #855 db021f1 Thanks @PuruVJ! - @ogygia/federation: a shell's pages allow island code from every peer origin in federate({ peers }), replicas included, so peer islands still wake under core's code-origin check (#851).
  • #703 1865089 Thanks @PuruVJ! - @ogygia/federation: an exposed route table is always asked on the app's own origin. A fragment request whose path does not start with exactly one / answers 400, and search is always the query.
  • #707 d03edda Thanks @PuruVJ! - @ogygia/federation: a deferred remote hole's signature now covers its peer, kind, target and search as separate fields (message version 2). A hole URL minted before the update is refused like any other bad signature; pages rendered after the update carry new ones.
  • #441 352a522 Thanks @PuruVJ! - A peer's redirect target is escaped into the "moved" card's link, so a quote in it no longer breaks the attribute.
  • #1973 cf48d2f Thanks @PuruVJ! - A peer page that is not found is now handled one way: it is a 404 wherever the shell still sets the status, and a slot that can't change its status keeps its fallback. A deferred page hole now answers 404 and keeps its fallback (it used to answer 200 and put the peer's not-found page in the region). A streamed mount({ stream }) now shows the "temporarily unavailable" card (it used to put the peer's not-found page inside the 200). mount() and mount.kit() still answer 404, as before. A peer page that moves with a status that is not a redirect (such as 304) is now refused as a bad answer (a 502, federation_peer_answer_invalid) instead of being passed on as a redirect.
  • #776 702f88f Thanks @PuruVJ! - @ogygia/federation: a signed request's x-og-sig is accepted only as the canonical base64 of its 64 bytes, so every copy of a signature counts as the same one.
  • #441 352a522 Thanks @PuruVJ! - A mounted peer page's title is escaped once: "Tom &amp; Jerry" no longer shows as "Tom &amp;amp; Jerry" in the tab, search results or the error page.
  • #1503 778d273 Thanks @PuruVJ! - A federation request (/og/frag, /og/thaw, /og/fragment/*) that reaches a server before any federate() ran there is now said once as federation_unregistered, instead of a bare 404 with no word.
  • #1457 620d27e Thanks @PuruVJ! - The head's end is found as the browser reads it: an app.html that writes </HEAD> gets ogygia's head injections, and a </head> inside a head script's text (a JSON-LD string) no longer takes them into the script. The body's end is found in any case.
  • #938 f889cbf Thanks @PuruVJ! - A deferred peer region whose peer fails (an error status, a cut or malformed body, or no answer) no longer answers its hole as an empty success: the hole answers 502, 503 or 504, so the region keeps its fallback and the browser asks again.
  • #878 08fd293 Thanks @PuruVJ! - A mounted peer page's head is read as a browser reads it: a title or stylesheet inside a comment is ignored, an unquoted rel counts, and a meta whose quoted content holds > is carried whole instead of cut.
  • #1116 9f7eac5 Thanks @PuruVJ! - @ogygia/core: Reading tags out of server HTML (the head dedupe, region CSS, hole answers, seed injection, stitching) costs 2 to 53% fewer steps: each tag is read once, a character at a time. @ogygia/federation: Lifting a peer page and pinning its asset refs costs about half the steps: one walk, each tag's attributes read once.
  • #930 ad132e5 Thanks @PuruVJ! - A peer answer that is not its JSON document no longer turns the shell's page into a 500: it is said as federation_peer_answer_invalid, a static widget shows its boundary card, and a page read or a mutation is a 502.
  • #1104 e09c8ae Thanks @PuruVJ! - A peer answer that is JSON but not its document's shape ({}, a proxy's {"error":…}, a field of another type) is a failed hop like one that does not parse: federation_peer_answer_invalid is said, a static widget shows its boundary card and a page read is a 502 (before, a TypeError made the shell's page a 500).
  • #932 cd255e1 Thanks @PuruVJ! - A static widget whose peer answers an error status (4xx or 5xx) no longer turns the shell's page into a 500: it shows its boundary card, the answer is said as federation_peer_answer_invalid, and a page read or a post passes the peer's own 503 or 504 through (any other error status is still a 502).
  • #1373 f3fd33c Thanks @PuruVJ! - A peer region's endpoint is re-pointed through the shell however its tag is written (single quotes, no quotes, spaces around =, upper case, after a value holding >), and text that only looks like an endpoint inside another value is left alone.
  • #1750 d780369 Thanks @PuruVJ! - A mounted peer page no longer carries its ogygia-* metas into the shell's head: the peer's router meta turned on the client router in a shell built with router: false, and overrode a shell's own view-transition setting.
  • #2025 7582a45 Thanks @PuruVJ! - A peer page mounted in a shell keeps the context its page set (createContext's setter, ogygia's setContext): it reaches the peer's own islands, scoped to the mounted page.
  • #1821 8684e66 Thanks @PuruVJ! - A page read from a peer whose fetch a thaw or a mutation drops every time no longer fetches forever: after 16 tries it is a 504, said as the new warning federation_peer_read_dropped.
  • #919 9215339 Thanks @PuruVJ! - A server island (or any region with its own endpoint) inside a peer's fragment now shows: the fragment pins its endpoints to its origin, and the shell serves each as a hole of its own that forwards to the peer, instead of asking its own endpoint and getting a 403.
  • #858 8769852 Thanks @PuruVJ! - @ogygia/federation: a peer that serves its island code from its own assets origin (a CDN) wakes on the shell: the shell's page allows the origin of each peer fragment's own runtime, taken only from that peer's answer (#856).
  • #962 b4d8675 Thanks @PuruVJ! - A static widget whose peer gives no answer at all (down, unreachable, or past its timeout) no longer turns the shell's page into a 500: it shows its boundary card, and the failure is said as the new warning federation_peer_unreachable (also for a page read, a post and a deferred region, which answer 504).
  • #1012 26f7e95 Thanks @PuruVJ! - A deferred peer region (and any plugin capability) names the key that signed it, as a region hole does: a server of another deployment (a rolling deploy without one shared OGYGIA_SECRET) refuses it as foreign-key, with the advice that goes with it, instead of as a bad signature.
  • #1470 a13ad32 Thanks @PuruVJ! - A plugin's signed hole (federation's /og/frag) now pays the visitor's region rate budgets as a region hole does: the probe budget before the signature check, the render budget after. A federation hole can now answer 429 with Retry-After under load.
  • #992 edff1c1 Thanks @PuruVJ! - @ogygia/federation: a thaw notice whose id, hop or tags is not the notice's shape is refused with a 400, applies nothing and is passed on to no one (#988).

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/flags

Minor Changes

  • #715 9604456 Thanks @PuruVJ! - @ogygia/router, @ogygia/flags: Breaking: anonymousVisitor() returns { vid, anonymous: true } and never fills sub, which stays a signed-in user's id (so mount() never signs a cookie value onward as one). Flags hash on sub, else vid (an anonymous visitor keeps its bucket), the og-vid cookie reaches a source as { vid, anonymous: true }, and the default openfeature() / ofrep() context sends vid as targetingKey when there is no sub. Code that read c.visitor.sub for anonymous visitors reads c.visitor.vid; exposure events from them arrive without a sub.
  • #1363 c7ed8cf Thanks @PuruVJ! - The flag inventory (node_modules/.ogygia/flags-manifest.json) is now written by flags(): a build writes it only when flags() is listed beside ogygia(), and an app that imports @ogygia/flags without listing it gets a one-time plugin_not_listed build warning.

Patch Changes

  • #442 95ed95c Thanks @PuruVJ! - say() in @ogygia/machine says a thing once per key until it recovers, remembering at most 1024 keys; a machine's warn-once and a plugin's warn use it, so a long-lived server no longer keeps every distinct warning text.
  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #428 1c7580f Thanks @PuruVJ! - In dev, a flag deleted from a module is gone once that module reloads: a flag source is no longer asked about it, and its layer no longer counts it.
  • #1220 8cb03fd Thanks @PuruVJ! - A flag source that does not answer within 1 second no longer holds the request: its flags fall to their native rule for that request and the failure is said (flags_source_failed). openfeature() and a custom source had no deadline, so one stalled provider stalled every page.
  • #440 4a038b4 Thanks @PuruVJ! - A cached hole (maxAge) that reads a flag, or a cookie, header or locals value, for an anonymous visitor is no longer kept and served to other visitors.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/insights

Minor Changes

  • #527 a4f938a Thanks @PuruVJ! - - @ogygia/insights: new package, the analysis engine the devtools dock, the profiler's report and @ogygia/testing share (what the browser measured, turned into findings). - @ogygia/core: breaking: the @ogygia/core/insights import path is gone. Install @ogygia/insights and import from it instead; the names are the same.

Patch Changes

  • #1416 14955b8 Thanks @PuruVJ! - The page weight report reads the page's tags as the browser does: a comment ends where the browser ends it, and a tag written inside a <title> or <textarea> is no asset. The held-open report reads the page seed from its own script, not from the page facts script beside it.
  • #529 32177c7 Thanks @PuruVJ! - - @ogygia/core: four new runtime hooks for a plugin's browser half: regionWoke, regionMismatched, navStarted and loadStarted; navigated now says whether the regions were reconciled, and holeFailed carries the hole's el. - @ogygia/devtools: the dock's Page, Hydration and Session readers and the testing API read the runtime hooks instead of core's event layer. - @ogygia/insights: a recorded session holds the runtime hook calls (SessionData.hooks, HookEntry) instead of core's events.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/jumpstart

Minor Changes

  • #1895 98b8e6a Thanks @PuruVJ! - New package. jumpstart() from @ogygia/jumpstart, listed beside ogygia(), builds the server in a few files for a fast cold start: dependencies bundled in, the framework in one chunk and each region registry in its own, an external option for packages that cannot be bundled, and one build line naming what stayed outside.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/machine

Minor Changes

  • #442 95ed95c Thanks @PuruVJ! - say() in @ogygia/machine says a thing once per key until it recovers, remembering at most 1024 keys; a machine's warn-once and a plugin's warn use it, so a long-lived server no longer keeps every distinct warning text.
  • #445 0a1e115 Thanks @PuruVJ! - flights() carries a priority: get(input, { priority }) raises a flight in progress to the highest priority of the callers that joined it (never lowers it), and the run reads flight.priority and hears each raise through flight.onRaise().
  • #416 34420d0 Thanks @PuruVJ! - flights() now keeps at most 1000 answers by default (the least recently used go first); pass max to change it. A cache() or flights() max must be a finite number.
  • #416 34420d0 Thanks @PuruVJ! - New cache(): a keyed store that is always bounded (max, weigh, keep), owned by a scope, and says why it lets each value go (onEvict); flights() keeps its answers in one.
  • #769 0ba97e8 Thanks @PuruVJ! - machine: capability() makes signed, expiring tokens addressed to an audience: length-prefixed messages that name the capability and its version, a lifetime clamped at mint and refused past its bound at check, a random nonce with optional one-time use in bounded memory, and no token from a clock that reads NaN; the signer (HMAC, Ed25519) is the caller's, and a check may bring its own verify (keys in rotation); skew allows for two machines' clocks; nonce: false mints the same token for the same inputs (not single-use, so replay is refused with it)
  • #473 382525a Thanks @PuruVJ! - New clock (clock.mono() for durations, clock.wall() for timestamps). A flights() keep and a retry policy() deadline are now measured on the monotonic clock, so a wall-clock step (an NTP correction, a VM resume) no longer shortens or stretches them.
  • #1935 8cf2b7d Thanks @PuruVJ! - New decisions(): a pure decision table whose inputs each take a listed set of values, so its whole input space can be walked. decide() throws on a combination refuse names (machine_decision_refused), and in dev on a value outside its list (machine_decision_outside). The test-only entry @ogygia/machine/testing walks a table: rows, check (named rules, each broken one named with its first row), table (markdown), dead_fields and count, with a 100,000-row guard (machine_decisions_too_many) and seeded sampling.
  • #428 1c7580f Thanks @PuruVJ! - New defineRun(id) and defining(): a plugin's definer calls owned by the run of the module that made them, so a definition deleted from the code leaves at the module's dev reload.
  • #476 c4f48c3 Thanks @PuruVJ! - New noop export: the one empty function, for a callback or cleanup with nothing to do.
  • #464 6d0a5cb Thanks @PuruVJ! - New queue(): a bounded, first-come queue. At most max jobs hold a place; a release() hands its place to the next job in the same step. A job can carry a deadline (all deadlines share one timer) or be cancelled, and the queue's owner closing makes the waiting jobs stale.
  • #416 34420d0 Thanks @PuruVJ! - New registry() and registry.keyed(): owned entries and definitions that leave with their owner, never evicted; cache() takes pinned(value, key) for values in use that never go for size.
  • #478 59fef58 Thanks @PuruVJ! - New slot({ name, version, make }): one process-global value per name, shared by every copy of a package a process loads. The version lives only in version (a name carries none): another version is refused in dev and kept apart in production.
  • #672 e3b6b5f Thanks @PuruVJ! - trace() in @ogygia/machine writes every machine move as one short line (visit#3 loading -> answered), the last 500, for a test to print when it fails.
  • #1839 4d0d5bb Thanks @PuruVJ! - queue() serves lines with more rules than first come: priority(job) (read again each pass, ties in the order pushed), ready(job) (a job not ready is passed over, the rest still start), windows: [{ max, holds }] (caps over some jobs, beside max), budget: { ms, yield, defer?, settle? } (a long drain hands the thread back), and recheck(). max is now optional, and a budget yield that throws is reported (machine_queue_yield_failed).
  • #1234 b97908b Thanks @PuruVJ! - queue() hands each started job its own place: start(job, place), given back with place.release(), once (a second release frees no other job's place). queue.release() is gone. A start that throws now gives its place back, so the jobs after it still get one.
  • #1842 b6b930e Thanks @PuruVJ! - These methods are now typed as function properties, so passing one on by itself (const { handle } = router) type-checks under unbound-method; none of them used this. A method type was bivariant in its parameters and a function property is checked strictly, so code that assigned a narrower function to one of them may now need its parameter types widened. The changed types: ogygiaTransport.Region.encode / .decode and the ref entry's encode / decode (@ogygia/core), CapabilitySigner.sign / .verify (@ogygia/core), Router.handle (@ogygia/router), Signer.sign / .verify and Staging.discard (@ogygia/machine).

Patch Changes

  • #1027 1b69b9f Thanks @PuruVJ! - @ogygia/machine: Faster hot reads on cache() and registry(): a get of the value set or read last no longer moves or looks it up again (a kept flights answer read by every island takes about half the steps), a cache with no onEvict queues no notices, observers are walked from one array kept until they change (registry().snapshot()), and registry.keyed() walks its definitions without a generator.
  • #1029 40b91ab Thanks @PuruVJ! - @ogygia/machine: A cache() set of the value its key already holds keeps it (no 'delete' notice, so an onEvict cleanup no longer runs on a value still held), set takes its own keep, and a keyed flights of one field keys by that field's value with no string built. @ogygia/core: A region answer written to the frame store costs about a third fewer steps (one cache write, no lookup before it), and an island warm reads its chunk graph once.
  • #419 1e7c5d4 Thanks @PuruVJ! - observe() keeps its watchers in a registry: each leaves with its stop, and the rest hear moves in the order they started watching.
  • #902 c9d5323 Thanks @PuruVJ! - - @ogygia/machine: cache() and registry() are smaller in a bundle (about 240 B brotli off a page's boot), and a bundle that uses registry() no longer carries defineRun() and defining().
  • #1017 5c16332 Thanks @PuruVJ! - @ogygia/machine: queue() with no now reads its deadlines on clock.mono(), like every other primitive, so a run that drives the clock drives the queue too.
  • #778 9b9f253 Thanks @PuruVJ! - core: region links are signed through capability(), naming the islands endpoint they are for and refusing a lifetime longer than the app ever mints; links stay identical between renders, and links signed before (pages cached before the upgrade) still work and are re-signed in the new form when renewed machine: capability().mintUntil() mints a token that expires at a given time, so tokens without a nonce stay identical across a window the caller picks
  • #1236 7c9ef65 Thanks @PuruVJ! - @ogygia/machine: in production, a slot() asked for at a version other than the one held keeps one private value for that version, made once. It made a new value on every ask, so state kept in it was lost.

@ogygia/postrender

Minor Changes

  • #1665 096de6c Thanks @PuruVJ! - Breaking: postrender() requires dataShake(). Add import { dataShake } from '@ogygia/data-shake/vite' and list dataShake() beside postrender(). Without it the build fails with plugin_needs_plugin, because every deferred region would look the page up and render the stored page again.
  • #747 771fa53 Thanks @PuruVJ! - @ogygia/postrender: Breaking: a page rendered for a request that carries a cookie or an authorization header is not stored (it is still served a stored copy, and its own render is stamped private, no-store); a route whose pages are the same with them opts in with config = { postrender: 'with-credentials' }. @ogygia/core: a render's per-visitor reads also count a function in locals, in and the keys of locals, the request headers read whole or copied while one is personal, headers.has() and a request clone(), client address, forwarded host, location and session headers, getClientAddress(), per-request platform data, and an own-origin fetch that carries the visitor's cookies.

Patch Changes

  • #475 81a40ea Thanks @PuruVJ! - Durations are measured on the monotonic clock: the memory store's TTLs, a federation deploy check's retry wait, the profiler's recording window, deadlines and busy waits, the testing API's wait, and the perf CLI's waits. A wall-clock step no longer moves them.
  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #949 dc3363f Thanks @PuruVJ! - A deploy now counts as a write for the app's own stored pages: on the first request after a boot with a new build, the pages the previous build stored are dropped (origin and edges) before any is served, so a new release never serves the old release's HTML.
  • #1457 620d27e Thanks @PuruVJ! - The head's end is found as the browser reads it: an app.html that writes </HEAD> gets ogygia's head injections, and a </head> inside a head script's text (a JSON-LD string) no longer takes them into the script. The body's end is found in any case.
  • #440 4a038b4 Thanks @PuruVJ! - A cached hole (maxAge) that reads a flag, or a cookie, header or locals value, for an anonymous visitor is no longer kept and served to other visitors.
  • #878 08fd293 Thanks @PuruVJ! - Stitching finds stitch holes as a browser reads the page: a hole inside a comment, or a region whose class merely says "stitch", is no longer stitched, and an upper-case or single-quoted stitch attribute is.
  • #1205 c9cbcc7 Thanks @PuruVJ! - A page whose store step fails after postrender read it (a stitched hole whose endpoint is not a URL, say) now goes out as rendered, no-store and unstored, and the failure is said once per page (postrender_capture_failed). The visitor got a response whose body was already read.
  • #1297 149157b Thanks @PuruVJ! - In dev, the would-store header and the "stays per-request" note now give the verdict production gives: a page the app marks no-store or private is no longer shown as would-store, and a permanent redirect is.
  • #1110 3d10f29 Thanks @PuruVJ! - A stored page answers If-None-Match with the weak comparison RFC 9110 asks for: a W/ etag a compressing CDN made, and *, now revalidate with a 304 (before, only the exact strong etag did, so such revalidations sent the whole page).
  • #955 79d90e7 Thanks @PuruVJ! - A remote function call (a query or query.live) made from a stored page is no longer answered with the stored page's HTML, and is never stored as the page: postrender stores and serves page requests only.
  • #1003 20ea2c3 Thanks @PuruVJ! - A stored page's stitch holes are found as the browser reads the page: a region tag inside a fallback's comment, script, style, title or textarea text no longer shifts which part of the page a hole covers, so an edge (ESI) rewrite never wraps half a region.
  • #963 a9fb365 Thanks @PuruVJ! - A stored page is served only to the build that stored it, recorded in the new optional build field of the public PostrenderEntry type (a custom store keeps it as part of the entry): another build's page is a miss, so a rolling deploy's old instances can no longer hand the new release their pages, and pages stored before this release are rendered again once.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/rewrite

Minor Changes

  • #1707 b20ae04 Thanks @PuruVJ! - - breaking: an app's own pass over the HTML ogygia sends is the new @ogygia/rewrite package. ogygia.handle({ transform }) now throws handle_transform_moved; list rewrite() (from @ogygia/rewrite/vite, no options) beside ogygia(), and put a rewrite() handle before ogygia.handle() in sequence(). ogygia/markup is removed, and scanRegions / liftRegions / restoreRegions leave ogygia/server. - @ogygia/rewrite: rewrite({ match, render }) renders only the outermost custom elements match picks (a tag prefix, a list of tags, or a test), in one render(elements, ctx) call per chunk of the page and per region answer, so a streamed page keeps streaming and its later chunks are rendered too. - @ogygia/rewrite: each element render gets has ogygia's marks numbered from 0, so a cache can key on its bytes; equal elements in one call are rendered once. - @ogygia/rewrite: ctx carries kind, csr, url, event and signal, an AbortSignal that fires when the request is cancelled. - @ogygia/rewrite: a render's sheets go out once per key per response, into the head while it is open and inline before the first element that needs them after. - @ogygia/rewrite: a failed batch is retried one element at a time; an element that still fails, drops ogygia's marks, or returns HTML that does not close what it opens goes out as ogygia rendered it, with a warning in every build (rewrite_element_failed, rewrite_marks_lost), never thrown. - @ogygia/rewrite: an element split across streamed chunks goes out unrendered, said in dev (rewrite_element_split). - @ogygia/rewrite: several rewrites in sequence() run in order on one marking of the HTML; one listed after ogygia.handle() throws rewrite_after_ogygia, and a rewrite without the plugin throws rewrite_plugin_missing. - @ogygia/rewrite: rewrite({ page }) rewrites the whole document (held until its last chunk) or region answer, says rewrite_page_failed when it throws (thrown in dev), and has .html(str) for a plain string. - @ogygia/rewrite: .check(html, { kind, csr }) runs the whole round trip on one piece of markup for a test, listing failures and, with a DOM, every restored element that differs from Svelte's. - @ogygia/rewrite: a page where nothing was restored ships no restorer, and each streamed chunk that restored something ends with one short restore call. - @ogygia/devtools: the profiler reads a page's regions itself and no longer depends on @ogygia/rewrite.

Patch Changes

  • #608 67127a4 Thanks @PuruVJ! - - @ogygia/core: the runtime no longer carries the restore support for an app's own HTML pass (the restore report, a web component's class tokens through hydration, the answer restore hooks); apps without @ogygia/rewrite boot about 0.15 to 0.2 kB (brotli) smaller. - @ogygia/rewrite: rewrite() ships that support to the browser itself; restore_markup_broken and restore_late are its message codes.
  • #855 db021f1 Thanks @PuruVJ! - @ogygia/rewrite: the restorer's attribute reset (og-r) never sets an on* or srcdoc attribute, or a javascript: / vbscript: URL (#851).
  • #1681 3398571 Thanks @PuruVJ! - A restored host's shadow root is now declarative, as a server-rendered <template shadowrootmode> root would be: a component whose constructor attaches its own root no longer throws on upgrade, and gets that root back emptied. A host not yet upgraded is replaced by an equal element during the restore, before anything hydrates.
  • #1693 29ac140 Thanks @PuruVJ! - A marked host that cannot have a shadow root no longer stops the restore: it is left as served with ogygia's marks taken off, said once per page and element name (restore_host_invalid), and every other host is still restored.
  • #1886 1b756bf Thanks @PuruVJ! - The restorer refuses a data: URL in an attribute reset (og-r), as it refuses javascript: and vbscript:, and reads each value's scheme the way a browser's URL parser does (the same check as core's html).
  • #1679 aa32050 Thanks @PuruVJ! - A restored host whose og-shadow names a sheet the page does not carry is no longer skipped silently: the page says restore_sheet_missing once per key, and the other sheets it names still apply.
  • #710 28f8504 Thanks @PuruVJ! - @ogygia/rewrite: the restorer for an app's server transform turns an <og-as> stand-in back only into a tag the server pass writes as one (a block, a list or table part, a link, a button, a form, an option), and never copies an on* attribute onto it. Any other <og-as> is left as it is.
  • #720 47d4d47 Thanks @PuruVJ! - The restorer reshape inlines into a document the app's transform reshaped is now written the same safe way as every other inline script (no text in it can end the script element early).
  • #1393 9872ace Thanks @PuruVJ! - The reversible transform's marking reads the page as the browser does: a quote inside an unquoted value or an attribute name no longer hides every custom element after it, a comment ends at --!> and <!--> too, and a custom element inside a <noscript> is left as the text it is.
  • #1832 b9edaf3 Thanks @PuruVJ! - A matched element whose start tag a chunk boundary cuts, after a chunk with no match, is now said as split (rewrite_element_split, in dev) like any element its chunk ends inside, instead of going out unrendered with nothing said.
  • #1830 abdfed1 Thanks @PuruVJ! - A matched element inside a declarative shadow root the page opened, in a later chunk of a streamed page, is now restored: that chunk's restore call enters every shadow root.
  • #1779 fa587dc Thanks @PuruVJ! - A stand-in tag whose attributes include a name only the HTML parser accepts (like =eq) is restored to its real tag, and the restore goes on to every stand-in and host after it instead of stopping with an error.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/router

Minor Changes

  • #715 9604456 Thanks @PuruVJ! - @ogygia/router, @ogygia/flags: Breaking: anonymousVisitor() returns { vid, anonymous: true } and never fills sub, which stays a signed-in user's id (so mount() never signs a cookie value onward as one). Flags hash on sub, else vid (an anonymous visitor keeps its bucket), the og-vid cookie reaches a source as { vid, anonymous: true }, and the default openfeature() / ofrep() context sends vid as targetingKey when there is no sub. Code that read c.visitor.sub for anonymous visitors reads c.visitor.vid; exposure events from them arrive without a sub.
  • #1143 c9dc308 Thanks @PuruVJ! - @ogygia/router: Breaking: when route patterns overlap, a path now resolves to the route SvelteKit would pick for the same files (an [[optional]] that is not last counts for nothing, so /[[lang]]/docs ranks as /docs; /b beats /b/[[page]]; ties go by route id, never table order), and CompiledPattern no longer has a score field.
  • #1842 b6b930e Thanks @PuruVJ! - These methods are now typed as function properties, so passing one on by itself (const { handle } = router) type-checks under unbound-method; none of them used this. A method type was bivariant in its parameters and a function property is checked strictly, so code that assigned a narrower function to one of them may now need its parameter types widened. The changed types: ogygiaTransport.Region.encode / .decode and the ref entry's encode / decode (@ogygia/core), CapabilitySigner.sign / .verify (@ogygia/core), Router.handle (@ogygia/router), Signer.sign / .verify and Staging.discard (@ogygia/machine).

Patch Changes

  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #1142 2ea1252 Thanks @PuruVJ! - href(), api() and entries() write a route's URL as Kit does: a [...rest] value keeps its slashes (each segment encoded, no more %2F), and an empty [[optional]] or [...rest] segment is dropped (/a/b, not /a//b). A rest value with an empty piece (a leading, trailing or doubled /) or a ./.. piece throws, and so does a param that fills a segment other than the last one empty, so a filled URL never names another origin or another route.
  • #1457 620d27e Thanks @PuruVJ! - The head's end is found as the browser reads it: an app.html that writes </HEAD> gets ogygia's head injections, and a </head> inside a head script's text (a JSON-LD string) no longer takes them into the script. The body's end is found in any case.
  • #878 08fd293 Thanks @PuruVJ! - A streamed page whose head closes as </head > (space or newline before >) gets its late boot, and a comment naming </head> before the page no longer takes it.
  • #1100 4dfe7a7 Thanks @PuruVJ! - api(), href() and entries() throw on a param of . or .. that fills a whole path segment, instead of building a URL that names another route (/users/[id]/delete with id: '..' asked /delete).
  • #700 ff753ff Thanks @PuruVJ! - @ogygia/router: a route segment with several params ([a]-[b]-[c]) matches in time in line with the path, and splits as SvelteKit's lazy params do (/x-y-z on [a]-[b] gives a: 'x', b: 'y-z'; /p-q-x-r on [a]-x-[b] gives a: 'p-q', b: 'r'). Paths longer than 8192 characters match no route.
  • #1186 033654a Thanks @PuruVJ! - A route load that fails runs once per request: the error page no longer runs it again to find the failed layout (its side effects ran twice, and a flaky load that passed the second time put the error page in the wrong layout). A later call to the same load in that request gets the same failure.
  • #1760 770cbb4 Thanks @PuruVJ! - c.json() and c.text() keep the headers they are given as a Headers object or as pairs, and a Content-Type you pass (in any case) replaces the default instead of joining it.
  • #701 a0a50ed Thanks @PuruVJ! - @ogygia/router: the trailing-slash redirect (slash: 'never' | 'always') always points at a path on this origin: its Location starts with exactly one / and holds no \.
  • #1227 69bd68d Thanks @PuruVJ! - A form POST naming an action the page does not have (?/delete on a page with only default) now answers 404, as in Kit. It ran the default action instead.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/testing

Minor Changes

  • #603 cb841aa Thanks @PuruVJ! - OgygiaDriver gains an optional addInitScript(source): new Ogygia(driver) hands it the testing engine as it is made, so a custom driver (Puppeteer's evaluateOnNewDocument, for one) gets the engine in every page it loads after that. playwrightDriver and withOgygia use it.
  • #531 dc985cf Thanks @PuruVJ! - - @ogygia/testing: withOgygia injects its own copy of the testing engine into each page before it loads (one bundled script, Playwright's addInitScript), so the engine always matches the package's version. page().self.bytes is 0 then (the page downloads none of it), and page().self.phases.engine is the engine's own start. - @ogygia/devtools: the dock's realm lends the injected engine what only the app's build knows (dev or not, the regions' names, the props lookup), and the leftovers watch keeps one record per page that any copy of the readers can read.
  • #605 ec75d61 Thanks @PuruVJ! - - @ogygia/testing: testing() from @ogygia/testing/vite, listed beside ogygia(), feeds the testing engine a test injects, so page(), hydration(), record() and settled() work without devtools in the app. A visitor runs nothing of it. profile() still needs devtools({ profiler }). - @ogygia/testing: the injected engine starts the page beacon itself, so a page without the devtools server is measured too. - @ogygia/devtools: with both the dock and testing() in an app, each runtime hook call is recorded and measured once.

Patch Changes

  • #475 81a40ea Thanks @PuruVJ! - Durations are measured on the monotonic clock: the memory store's TTLs, a federation deploy check's retry wait, the profiler's recording window, deadlines and busy waits, the testing API's wait, and the perf CLI's waits. A wall-clock step no longer moves them.
  • #1192 b8d8cf0 Thanks @PuruVJ! - profile() asks the profiler with a POST (the driver's post), as the profiler's /page now takes a POST only.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

ogygia

Minor Changes

  • #508 eae1add Thanks @PuruVJ! - Breaking: the barrel rewrite is a plugin now, barrels() from @ogygia/barrels, listed beside ogygia() (plugins: [ogygia(), barrels(), sveltekit()]) with the same options. Use it in place of ogygia({ barrels }), which is a build error (barrels_option_removed) that names it; debarrel(), DebarrelOptions and DebarrelMatcher are gone from ogygia/vite.
  • #532 b011a71 Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking (internal paths only): ogygia/internal/region-core is gone, and ogygia/internal/compiler no longer exports set_parser or takes a pathModule option; they existed only to run the compiler in a browser. Import region and isRegion from ogygia. Build output is unchanged.
  • #1665 096de6c Thanks @PuruVJ! - Breaking: shipping only the page.data your islands read is a plugin now, dataShake() from @ogygia/data-shake/vite, listed beside ogygia() (plugins: [ogygia(), dataShake(), sveltekit()]). Without it every island that reads the page gets all of page.data, and every server hole looks the page up.
  • #610 e07f6e4 Thanks @PuruVJ! - breaking: ogygia depends only on @ogygia/core and @ogygia/compiler, and no ogygia/* path forwards to a plugin package. ogygia/insights is removed: install @ogygia/insights and import from it (same names).
  • #693 b0a4f0e Thanks @PuruVJ! - region(C, props, { version }) in @ogygia/core: when two answers for the same held region reach the browser (two commands that both return it), the one with the newer data version stays on screen, whichever arrived last.
  • #528 9201d45 Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking: the internal internal/compiler-browser and internal/compiler-browser-full import paths are gone. The compiler runs in Node only; use @ogygia/core/internal/compiler there. - @ogygia/cli: breaking: the ogygia mcp server no longer has the ogygia_observatory tool; the Observatory it linked to is gone. - @ogygia/content: its macros and markdown pipeline read the compiler through @ogygia/core/internal/compiler and Node's own fs, path and crypto.
  • #1707 b20ae04 Thanks @PuruVJ! - - breaking: an app's own pass over the HTML ogygia sends is the new @ogygia/rewrite package. ogygia.handle({ transform }) now throws handle_transform_moved; list rewrite() (from @ogygia/rewrite/vite, no options) beside ogygia(), and put a rewrite() handle before ogygia.handle() in sequence(). ogygia/markup is removed, and scanRegions / liftRegions / restoreRegions leave ogygia/server. - @ogygia/rewrite: rewrite({ match, render }) renders only the outermost custom elements match picks (a tag prefix, a list of tags, or a test), in one render(elements, ctx) call per chunk of the page and per region answer, so a streamed page keeps streaming and its later chunks are rendered too. - @ogygia/rewrite: each element render gets has ogygia's marks numbered from 0, so a cache can key on its bytes; equal elements in one call are rendered once. - @ogygia/rewrite: ctx carries kind, csr, url, event and signal, an AbortSignal that fires when the request is cancelled. - @ogygia/rewrite: a render's sheets go out once per key per response, into the head while it is open and inline before the first element that needs them after. - @ogygia/rewrite: a failed batch is retried one element at a time; an element that still fails, drops ogygia's marks, or returns HTML that does not close what it opens goes out as ogygia rendered it, with a warning in every build (rewrite_element_failed, rewrite_marks_lost), never thrown. - @ogygia/rewrite: an element split across streamed chunks goes out unrendered, said in dev (rewrite_element_split). - @ogygia/rewrite: several rewrites in sequence() run in order on one marking of the HTML; one listed after ogygia.handle() throws rewrite_after_ogygia, and a rewrite without the plugin throws rewrite_plugin_missing. - @ogygia/rewrite: rewrite({ page }) rewrites the whole document (held until its last chunk) or region answer, says rewrite_page_failed when it throws (thrown in dev), and has .html(str) for a plain string. - @ogygia/rewrite: .check(html, { kind, csr }) runs the whole round trip on one piece of markup for a test, listing failures and, with a DOM, every restored element that differs from Svelte's. - @ogygia/rewrite: a page where nothing was restored ships no restorer, and each streamed chunk that restored something ends with one short restore call. - @ogygia/devtools: the profiler reads a page's regions itself and no longer depends on @ogygia/rewrite.
  • #516 368b1da Thanks @PuruVJ! - - @ogygia/core, ogygia: breaking: SharedState is no longer exported from ogygia or @ogygia/core. Import it from @ogygia/federation/shared instead; it works the same. - @ogygia/federation: SharedState (state shared across separate builds on one page) is exported from @ogygia/federation/shared.

Patch Changes

  • #533 49ff074 Thanks @PuruVJ! - - @ogygia/compiler: new package, ogygia's compiler and its Vite plugin: @ogygia/compiler (the compiler), @ogygia/compiler/vite (ogygia(), definePlugin) and @ogygia/compiler/ids. It uses core only through @ogygia/core/contract; core never depends on it. - @ogygia/core: breaking: the @ogygia/core/vite, @ogygia/core/internal/compiler and @ogygia/core/internal/compiler-ids import paths are gone. Import from ogygia/vite (or @ogygia/compiler/vite), @ogygia/compiler and @ogygia/compiler/ids instead; the names are the same. - @ogygia/core: the types of the ogygia:* virtual modules and the build's switches are named in core (@ogygia/core/contract), so core's own types no longer reach into the compiler. - ogygia: ogygia/vite, ogygia/internal/compiler and ogygia/internal/compiler-ids re-export @ogygia/compiler; apps change nothing. - @ogygia/barrels, @ogygia/cli, @ogygia/content, @ogygia/devtools, @ogygia/federation, @ogygia/flags, @ogygia/postrender, @ogygia/router: they take definePlugin and the compiler from @ogygia/compiler, a new peer dependency (ogygia brings it).
  • #576 7be0138 Thanks @PuruVJ! - The head's duplicate-link pass reads each <link> attribute as HTML does, so a stylesheet or preload hint whose URL has an apostrophe is no longer dropped, and data-rel / data-href no longer count as rel / href.
  • #878 08fd293 Thanks @PuruVJ! - A hole answer's region URLs, island graph and region CSS are read as a browser reads the tags: a region or sheet inside an HTML comment is left alone, and upper-case tags, single-quoted values and a quoted > no longer hide one.
  • #577 13aa062 Thanks @PuruVJ! - A $page.data promise nested more than 65 levels deep, or inside a cyclic value, now streams into regions like any other: before, it was left in the seed, the page failed to serialize it, and a rejection went unhandled.
  • #577 13aa062 Thanks @PuruVJ! - On a router (non-navigate) request, $page.data with promises keeps its key order in the seed; before, keys came in the order their values settled.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

0.13.0 minor

@ogygia/cli

Patch Changes

  • #264 bfa038f Thanks @PuruVJ! - ogygia bisect now measures the server it started: a commit's server is stopped and gone (SIGKILL after a grace time) before the next one starts, a port another process holds is refused as cli_server_port_busy, and a server that exits before it answers fails at once with its output as cli_server_exited_early (#250).
  • #305 e605070 Thanks @PuruVJ! - ogygia mcp loads the profiler's report reading again after a failed load, so installing @ogygia/devtools needs no restart (#295).
  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.
  • #340 24c6592 Thanks @PuruVJ! - ogygia_profile no longer takes over someone else's live recording: the profiler's 409 says who holds the recorder and whether that holder is stale, and the tool resets only a stale one; a live one gets a "busy" message (#290).

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/content

Minor Changes

  • #177 d9e8e2f Thanks @PuruVJ! - The headless on-device search handle is a class: new SearchClient({ base }) replaces search({ base }) (same options, same ready, query() and destroy()).

Patch Changes

  • #370 9272db9 Thanks @PuruVJ! - The code-block copy button no longer leaves a refused clipboard as an unhandled rejection, and a second copy within the tick keeps the tick for its full time (#365).
  • #318 085cb73 Thanks @PuruVJ! - Content's dev reload lets go of a change it was still waiting on when its dev server closes or restarts.
  • #257 c0b92c2 Thanks @PuruVJ! - An edit in dev no longer leaves another live connection open: a collection its module's next run replaces closes its connection, clears its retry and leaves the graph (#247).
  • #257 c0b92c2 Thanks @PuruVJ! - A live list or live get streaming a collection that a dev edit replaced now ends, instead of waiting forever for a change that cannot come (#247).
  • #220 8707255 Thanks @PuruVJ! - A dimension coordinate whose outline failed to build once builds it again on its next page (#192).
  • #220 8707255 Thanks @PuruVJ! - One bad live update no longer stops live updates: the entry keeps its last good version, and a dropped live connection is opened again (after a doubling wait) and the whole source read again (#190).
  • #220 8707255 Thanks @PuruVJ! - A change made while a collection's first read runs, or between a live list's first value and its next pull, is no longer missed (#191).
  • #371 0c1e449 Thanks @PuruVJ! - A dropped live connection is opened again under a named policy whose waits are spread (cut by up to a fifth), so every instance a source dropped at once no longer reconnects at the same moment.
  • #257 c0b92c2 Thanks @PuruVJ! - A markdown highlighter or mdsvex start that fails once (a grammar or plugin that failed to load) no longer breaks markdown until a restart: the next render starts it again (#248).
  • #220 8707255 Thanks @PuruVJ! - The site's outline (nav, pages, addresses) and its search index follow a live source's changes; a change the outline cannot be woven from keeps the last good outline and says why, and a failed search build is not kept (#193).
  • #220 8707255 Thanks @PuruVJ! - Every site search query is answered or rejected: a failed query, a failed index load, a worker crash or destroy() no longer leaves one waiting (#207).
  • #305 e605070 Thanks @PuruVJ! - The search worker downloads its index again for the next query after a failed download, instead of failing every query with the first error.
  • #318 085cb73 Thanks @PuruVJ! - The mobile bar's sheet lets go of its pinned height after a view switch with no glide (both views the same height), and on unmount.
  • #370 9272db9 Thanks @PuruVJ! - A Shiki theme object with no name now highlights (it is named from its contents), two such themes no longer share a highlighter or a fence-cache key, a theme object edited in place is a new fence-cache key, and snippets or alternating highlight() configs reuse one highlighter per set of themes and languages instead of making a new one each time (#366).
  • #318 085cb73 Thanks @PuruVJ! - The sidebar stops saving its panel's scroll position when it unmounts.
  • #257 c0b92c2 Thanks @PuruVJ! - A folder sidecar (+meta.json, index.md) that fails to load once no longer breaks the collection and its groups until a restart: the next read loads it again (#246).
  • #220 8707255 Thanks @PuruVJ! - A collection whose first load fails (the source down, its init or a schema check failing) no longer keeps the failure: the next read loads it again (#189).
  • #328 cc047cc Thanks @PuruVJ! - Two git loaders on one repo with another branch or folder each get their own checkout, instead of the second one finding the first one's files (or none) (#288).
  • #328 cc047cc Thanks @PuruVJ! - A markdown file moved to another folder with the same content is compiled again, instead of the markup cache handing it the old path's CSS key (#292).
  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.
  • #327 79aa5b3 Thanks @PuruVJ! - A live edit that lands while the outline is still reading a collection (its entries waiting on a relation) is woven in by the next read, instead of the outline from before it staying until the next edit (#326).
  • #398 0d76899 Thanks @PuruVJ! - Removed inline-marker text (---removed---) in a code fence now renders as escaped text, never as live HTML.
  • #305 e605070 Thanks @PuruVJ! - The search palette's download is asked for again after it fails, and a click while it cannot load follows the trigger's /search link (#272).
  • #305 e605070 Thanks @PuruVJ! - Ctrl/Cmd+K pressed again while the search palette loads takes the open back, so it stays closed (#279).
  • #321 17e009a Thanks @PuruVJ! - The search palette shows a failed query's error (and prints content_search_query_failed) instead of keeping the previous term's results up and leaving the rejection unhandled (#273).
  • #184 1767110 Thanks @PuruVJ! - Packages declare sideEffects (only their .css files run on import), so bundlers drop the modules an app never uses.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/core

Minor Changes

  • #174 413e77f Thanks @PuruVJ! - A page discloses which ogygia it runs on window.__ogygia.v (a Set of versions, as Svelte's window.__svelte.v), on csr=false and csr=true pages alike.
  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.

Patch Changes

  • #226 019b77e Thanks @PuruVJ! - In dev, a barrel saved while a transform resolves through it is built again: the build in flight stores nothing and its callers get the new export map, instead of imports rewritten to the old files until the next edit (#200).
  • #258 734c4c5 Thanks @PuruVJ! - Barrel imports in dev now resolve each leaf with the environment that asked: a leaf the client and the server resolve to different files (by conditions) no longer gets the other one's file, from a transform running at the same time or from the cached barrel map (#252).
  • #256 25ab8c7 Thanks @PuruVJ! - A batch hole stream whose visitor leaves now stops: the holes not rendered yet never render or take a render slot, and the answers that land after it are dropped instead of thrown into the closed stream (#245).
  • #316 da577ee Thanks @PuruVJ! - A batch of a page's holes stops when the visitor leaves the page (or when the navigation that started it is called off), and every way a batch ends lets go of its answer's body, the server's end mark included.
  • #223 ad8a29a Thanks @PuruVJ! - A batch stream that goes quiet for 10 seconds is given up: the holes it still owes fetch on their own instead of keeping their fallback forever (#214).
  • #225 29d8553 Thanks @PuruVJ! - The profiler's visit beacons no longer take the first tab switch for the page leaving: coming back to the tab or a back/forward cache restore opens the visit again, so later navigations, interactions and shifts go out with the next hide, and the standalone beacon sends no copy after its final visit (#209).
  • #256 25ab8c7 Thanks @PuruVJ! - Concurrent requests for one cached hole (a maxAge region) now share one render instead of each rendering it and taking a render slot; a failed render fails the requests that joined it and is not stored (#244).
  • #321 17e009a Thanks @PuruVJ! - An app's callback that throws (a beforeNavigate or afterNavigate hook, a page store subscriber, a ref watcher or disposer, a source receipt seam, a devtools sink) now reaches the page's error handler (reportError), or the console on a server, instead of being dropped; the others still run.
  • #265 ee66652 Thanks @PuruVJ! - A lazy chunk whose own dependency failed to download is asked for again once, not on every later wake: the runtime then says once that the page needs a reload (chunk_reload_needed), and a runtime that fails to load at all now says so too (runtime_load_failed) (#262).
  • #317 7f64cd1 Thanks @PuruVJ! - On a Kit-hydrated page, a custom element Kit removes (and mounts anew) leaves the class watch, and the watch's observer goes once no host is left, instead of holding the removed hosts for good (#281).
  • #186 01dc5b9 Thanks @PuruVJ! - Compiling a page with many regions is up to 1.9x faster: each region import no longer walks the whole file again to learn whether it is used.
  • #302 a39c26f Thanks @PuruVJ! - A waking region no longer searches the whole page for a script that needs devalue: the page is asked once, and again only when new markup comes in, so a 250-region page spends about 0.2 ms on it instead of about 48 ms.
  • #169 e8cd9f7 Thanks @PuruVJ! - A page whose islands get plain JSON props no longer downloads devalue: it loads only when a props sidecar, the page seed or page facts are in devalue's lane, beside the hydrate code (about 2 KB brotli off each island page of the benchmark).
  • #265 ee66652 Thanks @PuruVJ! - A click on an island before it is live now acts once, when it wakes: the island waits for what replays the click before it goes live, instead of letting the replay come in later, beside the visitor's next click (#260).
  • #256 25ab8c7 Thanks @PuruVJ! - An exposure sink that rejects no longer crashes the server: the failure is logged with a message code, and the sink's send is handed to the platform's waitUntil when it has one, so serverless instances no longer cut exposures off. The response never waits for the send (#243).
  • #384 4da8fe4 Thanks @PuruVJ! - In Firefox a chunk or island file whose download was dropped is asked for again under a fresh URL, as in the other browsers: a resource-timing status of 0 now reads as no response (#388).
  • #187 b46798f Thanks @PuruVJ! - The build's flag inventory takes time in line with a module's size: each flag's line is found from one pass over the file, not counted again from the top.
  • #340 24c6592 Thanks @PuruVJ! - A heading the router focused gives its tabindex back when focus leaves it, even after a second navigation focused it again meanwhile (a refresh in place at its #hash), or when it is kept across the navigation inside an island: it no longer stays in the tab order with tabindex="-1" for good (#278).
  • #253 b60df97 Thanks @PuruVJ! - A forced refresh at an address a navigation batch reserved no longer orphans the holes that joined the batch: they wait on the refresh, which settles them with the first frame written there or fails them with its own error, and a hole that binds after the batch ended fetches on its own (#233).
  • #253 b60df97 Thanks @PuruVJ! - A page's form restore window ends when the visitor goes on: fields restored on a page no longer leak into the next page's islands within 5 s (#235).
  • #324 7b35e33 Thanks @PuruVJ! - A router navigation after another one that brought Kit's FOUC <style> goes in as well, instead of throwing inside the head merge and loading the page whole.
  • #253 b60df97 Thanks @PuruVJ! - The newest frame a region gets paints last: an older live tick or hole refresh still waiting for its stylesheet no longer paints over a newer one that needed none (#231).
  • #305 e605070 Thanks @PuruVJ! - A region refresh that starts while a first fetch of the same hole is still out stops that fetch, and the regions that waited on it get the refreshed HTML.
  • #303 7a4bde1 Thanks @PuruVJ! - A hydrating region records its <head> content from what it added, not by copying and scanning the whole <head>: on a 250-region page with a large head this takes about 0.15 ms instead of about 58 ms.
  • #411 537c7f3 Thanks @PuruVJ! - A hole that loads at once is now asked for once in every browser, including Safari: before, Safari asked twice.
  • #339 f693f50 Thanks @PuruVJ! - A hole's island with wake: 'interaction' wakes on the visitor's first press or key when the hole's own when is another schedule; it used to be armed as the hole's on-demand fetch, which never fired, and stayed server markup for good (#338).
  • #223 ad8a29a Thanks @PuruVJ! - A hole's fetch, its answer and its island's wake are machines: an island wakes whoever fetched its answer, a kept fallback that later gets an answer wakes too, a failed refetch no longer marks the next answer a refresh, a region with two schedules drops both when it leaves, and an interaction island that fails to wake says so (#206).
  • #339 f693f50 Thanks @PuruVJ! - A hole put back on the page after a while off it shows its answer again and wakes its island, instead of showing nothing (its island had taken its markup with it) or keeping an island nothing wakes, once the frame store let the answer go (#336).
  • #223 ad8a29a Thanks @PuruVJ! - A renewed hole capability belongs to its address: every copy of the hole that took the answer moves onto it, not only the one that fetched (#213).
  • #223 ad8a29a Thanks @PuruVJ! - A copy of a hole that leaves the page gives up only its own place in the shared fetch: its twin's retry is no longer cancelled, and a twin that joined a batch no longer waits forever (#205).
  • #173 e7c31c0 Thanks @PuruVJ! - Islands hydrate the component itself instead of through a wrapper component: an island page ships less Svelte (about 0.3–0.4 KB brotli), and a placed island's server markup carries one marker pair fewer.
  • #253 b60df97 Thanks @PuruVJ! - A region put back while its island loads wakes once, and the old run's failure no longer reaches the new visit: no error logged against the new module, no failure report, its drift watch and scheduler place kept (#230).
  • #387 4a45161 Thanks @PuruVJ! - A region or hole answer is parsed in a document with no browsing context, so no engine builds its custom elements before the restore (Firefox 159 did, and the page showed a component's server render twice); a host a component runtime upgraded first keeps its own root, gets its light DOM back without the server's copy, and says restore_late in every build (#382).
  • #253 b60df97 Thanks @PuruVJ! - An early-click capture ends only its own arming: a region put back while its first wake loaded keeps catching clicks, and the old capture no longer replays its clicks onto the new visit (#232).
  • #265 ee66652 Thanks @PuruVJ! - An island whose own file fails to download is asked for again under a fresh URL, at once and by every later wake, like the runtime's lazy chunks: one dropped download no longer leaves the island, and every island of that file, dead for the tab (#261).
  • #305 e605070 Thanks @PuruVJ! - Two islands of one file whose download fails together share one retry, so the module loads once instead of twice (#271).
  • #333 865eb55 Thanks @PuruVJ! - The dev rule that no shared work keeps a failure now checks what is kept now: once that failure is dropped or replaced, it no longer reports it again (#330).
  • #317 7f64cd1 Thanks @PuruVJ! - On a Kit-hydrated page, the wait for Kit's page ends once no island waits for it: an island that leaves the page while it waits no longer keeps the wait going (its timers and its watch of Kit's page).
  • #317 7f64cd1 Thanks @PuruVJ! - On a global that refuses an accessor on Kit's page symbol, an island waiting for Kit's page now hydrates once Kit publishes it, instead of waiting for good.
  • #223 ad8a29a Thanks @PuruVJ! - The islands inside a remounted swr lake wake once its refresh lands, not only when the refresh fails (#204).
  • #321 17e009a Thanks @PuruVJ! - An inline island a client navigation created whose code fails to load now prints island_load_failed, instead of rendering nothing without a word.
  • #384 4da8fe4 Thanks @PuruVJ! - An island file that 404s is no longer asked for again under a fresh URL when the browser's timing entry for it lands late (Firefox on a busy machine): the loader waits for the entry, up to a second.
  • #215 a7ce10d Thanks @PuruVJ! - A lazy runtime chunk (the hydrate core, devalue, the interaction replay) that fails to download is asked for again under a fresh URL, so one dropped request no longer leaves every later island wake in the tab dead (#201).
  • #187 b46798f Thanks @PuruVJ! - Finding the images that downloaded beside a page's largest paint takes time in line with its images.
  • #223 ad8a29a Thanks @PuruVJ! - A live region is one machine: a tick that comes while its island hydrates no longer mounts a second island, turning static disposes the island, and turning interactive stops the static morph (#211).
  • #339 f693f50 Thanks @PuruVJ! - A live region put back on the page goes on in its last tick's mode: a static one follows the writes at its address again, an interactive one hydrates its island again, instead of both stopping until the next tick (#337).
  • #315 7e7c047 Thanks @PuruVJ! - @ogygia/core: draining the load queue reads each waiting download's class once instead of sorting the whole queue every time a download ends, so a big page spends less time picking what to fetch next, in the same order.
  • #339 f693f50 Thanks @PuruVJ! - A region that leaves the page before its load schedule fires starts nothing: no fetch of a hole's answer off the page (which then gave up and reported a failure), and no wake of its island (#335).
  • #360 64431ac Thanks @PuruVJ! - A region's mailbox runs any number of events queued in one turn on a flat stack, instead of overflowing it after a few thousand sync ones (#342).
  • #339 f693f50 Thanks @PuruVJ! - A region whose element leaves the page while one of its waits is queued behind another event ends that wait with STALE, instead of leaving it (and the wake or live tick around it) unsettled for good (#373).
  • #360 64431ac Thanks @PuruVJ! - A region's mailbox hands each handler a turn of its own: a wait from a handler that does not hold the turn (two waits at once, a timer it left behind) no longer ends another handler's turn and lets two events of one element run at once (#341).
  • #316 da577ee Thanks @PuruVJ! - A navigation called off before its swap (back pressed, a link to the page on screen) takes the next page's stylesheets it put in the head out again, and stops waiting for them at once, instead of leaving them to style the page on screen (#267).
  • #216 e788339 Thanks @PuruVJ! - Two links clicked while the router chunk downloads: the last one wins, not the first (#203).
  • #324 7b35e33 Thanks @PuruVJ! - The next page's stylesheets load inert while a navigation waits, and style the page only once it goes in: a sheet that arrived first no longer styles the page still on screen (#267).
  • #324 7b35e33 Thanks @PuruVJ! - A navigation whose swap throws (with or without a view transition), or whose devalue fails to load, now ends and goes to the browser with its error said, instead of staying in flight or finishing as if the page went in (#276).
  • #316 da577ee Thanks @PuruVJ! - The page facts a hole's answer brings are its page's: a navigation that swaps the next page in before any island woke no longer merges them over the next page's page.data (#269).
  • #298 28e957d Thanks @PuruVJ! - A value the page seed cannot carry (a function, a symbol, an unknown class) is left out where it sits, so regions still get the rest of page.data, form and error instead of undefined.
  • #370 9272db9 Thanks @PuruVJ! - A streamed $page.data promise whose value throws as it is staged (a getter that throws) now fails that promise's {#await} instead of keeping the response open for good; the other promises still stream and the stream closes (#367).
  • #321 17e009a Thanks @PuruVJ! - A page made for one browser (devtools events, the profiler's beacon) whose headers cannot be changed now prints personal_response_unmarked, instead of going out without private, no-store and without a word.
  • #221 a5d28b0 Thanks @PuruVJ! - A plugin's shared ask hook throws when no implementation answered and one threw, instead of answering undefined as if nobody knew (#198).
  • #321 17e009a Thanks @PuruVJ! - A plugin's done that throws, or whose work fails after the response, now prints plugin_hook_failed with the error, instead of nothing; the response still never fails on it.
  • #258 734c4c5 Thanks @PuruVJ! - A dev edit that runs a plugin's server half again no longer leaves its old run in the shared hooks: the plugins of the run before leave first, so notify reaches only the current ones and ask answers from them (#251).
  • #381 debb73b Thanks @PuruVJ! - The browser runtime no longer ships the server's render and stream deadlines: each named retry policy is kept only where its code reads it.
  • #216 e788339 Thanks @PuruVJ! - A prefetch answered after a remote mutation cleared the page cache no longer puts the page from before the mutation back (#208).
  • #253 b60df97 Thanks @PuruVJ! - An old prefetch no longer forgets a newer prefetch of the same page when its slot comes up after a cache clear, so a click on that page lifts the newer one instead of waiting behind every other guess (#234).
  • #223 ad8a29a Thanks @PuruVJ! - preload() of a hole fetches it the way the hole does: a keepFallback() answer keeps the fallback instead of wiping it, a page or a redirect is refused, and a renewed capability is kept (#212).
  • #370 9272db9 Thanks @PuruVJ! - A profiler mark() made before the page's beacon loads now waits for it however late it comes, instead of being dropped after 10 s; the beacon takes the waiting marks as it loads (#369).
  • #370 9272db9 Thanks @PuruVJ! - A file the static host could not answer once (a 5xx, a timeout, a network blip) is weighed again by the next profiler report instead of being left out for the life of the process, and reports weighing a file at once share one fetch (#362).
  • #170 6a63ecf Thanks @PuruVJ! - The island prop-mutation warning (a dev aid) no longer ships in production builds: its guard is created only in dev, so the bundler drops it (about 0.4 KB brotli per island page).
  • #340 24c6592 Thanks @PuruVJ! - A region stylesheet whose download failed is asked for again by the next answer that carries it, which waits on that download only: the region no longer stays unstyled for the rest of the tab, nor waits 5 s on a sheet that never loads (#274).
  • #340 24c6592 Thanks @PuruVJ! - A hole kept across a navigation keeps its stylesheet: a region sheet stays in the head while a region on the new page still uses it, instead of going with the head merge because the new page's server head does not list it (#275).
  • #305 e605070 Thanks @PuruVJ! - A cached hole whose first request leaves while its render waits for a slot keeps its place in the render queue for the requests still waiting on it, instead of being rendered again from the back of the queue; a render every request left never takes a slot.
  • #328 cc047cc Thanks @PuruVJ! - A cached region render is keyed by its region, props, session and page as separate fields, so two visitors whose props and session spell one string another way never share a render (#312).
  • #218 e6aed57 Thanks @PuruVJ! - Hole renders never run above the concurrency limit: a freed slot goes straight to the next waiter, and a render past its timeout gives its slot to the next hole, so renders stuck on a hung upstream never block the others (#195).
  • #218 e6aed57 Thanks @PuruVJ! - A hole answers within the render timeout, its wait for a render slot included, and the server logs when a render first outlives it (#195).
  • #173 e7c31c0 Thanks @PuruVJ! - A live region and a woken deferred island now adopt their server HTML: they were thrown away and re-rendered on the client every time they hydrated.
  • #321 17e009a Thanks @PuruVJ! - A request's owned work now stops at once when its answer is a HEAD or a page SvelteKit turns into a 304, instead of running until the visitor leaves (#320).
  • #258 734c4c5 Thanks @PuruVJ! - A router page's CSS in dev is no longer lost after one failed read of the router CSS registry: the next render reads it again (#249).
  • #265 ee66652 Thanks @PuruVJ! - A link the browser keeps the page on screen for (a file download, a 204) works on every click again, not only the first: the router's hand-off of a navigation to the browser now ends it, so a later click on the link no longer joins it and gets swallowed (#266).
  • #265 ee66652 Thanks @PuruVJ! - Back from a page the router handed to the browser (a csr=true Kit page, a page that is not ogygia's) now loads the page before it: the router's history entry for the visit is replaced by the browser's document instead of reloaded, so the router page's entry no longer belongs to the new document (#259).
  • #325 4b63071 Thanks @PuruVJ! - The hashchange the router fires for an app-managed fragment link has the address before the click as its oldURL (#277).
  • #325 4b63071 Thanks @PuruVJ! - The router writes a history entry only when a page goes in: download and 204 clicks add none, and a download sent as HTML (Content-Disposition: attachment) is saved, not swapped into the page (#297).
  • #325 4b63071 Thanks @PuruVJ! - The router keeps the app's own history.state on the entries it takes over (the one it starts on, an app's own pushState met on Back or Forward, one replaced in place): its id goes beside that state instead of over it (#375).
  • #325 4b63071 Thanks @PuruVJ! - The router keeps one record of the visitor's history, and every page left keeps its scroll: Forward after Back lands where the page was left, not at the top (#270).
  • #254 346c692 Thanks @PuruVJ! - The router's navigation code that fails to download is asked for again under a fresh URL: the next click, invalidateAll, invalidate or preloadData loads it, instead of links falling back to full page loads and refreshes failing until a reload, and a prefetch it fails for warms nothing instead of leaving an uncaught error (#238).
  • #305 e605070 Thanks @PuruVJ! - A navigation that was waiting on a prefetch when a remote mutation cleared the page cache fetches the page again instead of showing it as it was, a prefetch of a page a navigation is fetching joins that request, and a prefetch the clear ended gives its turn back without fetching.
  • #265 ee66652 Thanks @PuruVJ! - A link component that clicks its inner <a> again after the visitor's click no longer turns the router's navigation into a full load when that second click never reaches the document (a click event that stays in its shadow root, or an anchor the swap took out of the page): it joins the navigation the first click started (#263).
  • #329 c781aeb Thanks @PuruVJ! - In dev, ogygia checks its own rules after each navigation and machine move (one mount per region, the address bar matches the router's history, nothing of a left page stays open, no failure leaves without a code) and after each response (a request's work ends with it, no shared work keeps a failure), and says a broken one with its code; a production build ships none of them (#313).
  • #253 b60df97 Thanks @PuruVJ! - The hydration scheduler wakes at its earliest wait: a short wait for a region's viewport report set during a long hold no longer waits out the hold (#236).
  • #300 b10c054 Thanks @PuruVJ! - Handing out hydrate turns costs the same per region however many are waiting: a page with 250 load regions spends about a third of the time scheduling them, 2,500 regions about a twentieth.
  • #299 2a970fe Thanks @PuruVJ! - Matching copied records in region props to the page seed takes time in line with the props: a list page whose region gets copies of its products renders its seed references up to 3.5x faster.
  • #380 469d980 Thanks @PuruVJ! - A region's viewport snapshot is held by its lifetime with one cleanup instead of a scope of its own, so a hydrate turn costs about what it did before the scheduler moved onto scopes.
  • #216 e788339 Thanks @PuruVJ! - A soft refresh (invalidateAll) still in flight when the visitor navigates no longer merges the old page's head and data into the new one (#202).
  • #324 7b35e33 Thanks @PuruVJ! - invalidateAll during a navigation refreshes the page on screen, not the page the navigation is still bringing, and a navigation that goes in ends a refresh still out (#268).
  • #323 7a53a56 Thanks @PuruVJ! - A publish's thaw notices go through an outbox handed to the request's waitUntil (the new app.keepAlive), so a serverless instance that freezes after the response no longer loses them and peers no longer keep stale pages until their TTL (#286).
  • #321 17e009a Thanks @PuruVJ! - A peer whose store fails to apply a thaw notice now answers an error, so the sender sends it again, instead of answering done and deduping the retry; app.notify rejects once every hook ran when one failed (#287).
  • #384 4da8fe4 Thanks @PuruVJ! - In Safari an island file whose download failed is asked for again under a fresh URL, as in the other browsers, and its failure is reported as a failed load, not a failed hydration (#383).
  • #254 346c692 Thanks @PuruVJ! - A transportable class whose module fails to download is asked for again: the next island that carries it loads it under a fresh URL and decodes it, instead of every later island decoding without the class until a reload (#237).

Updated dependencies: @ogygia/[email protected]

@ogygia/devtools

Minor Changes

  • #177 d9e8e2f Thanks @PuruVJ! - The profiler's stores are classes: new MemoryStore(), new SqliteStore(), new RedisStore() and new PostgresStore() replace memoryStore(), sqliteStore(), redisStore() and postgresStore() (same arguments, same entry points).

Patch Changes

  • #225 29d8553 Thanks @PuruVJ! - The profiler's visit beacons no longer take the first tab switch for the page leaving: coming back to the tab or a back/forward cache restore opens the visit again, so later navigations, interactions and shifts go out with the next hide, and the standalone beacon sends no copy after its final visit (#209).
  • #187 b46798f Thanks @PuruVJ! - Finding a load's constant work takes time in line with the load's lines: the functions it also calls elsewhere are read from one pass.
  • #255 fc35032 Thanks @PuruVJ! - Turning devtools off in a build now takes the whole dock with it: the app, its observers and its Alt+O listener go, a dock still downloading never mounts, and turning it on again runs one dock (#239).
  • #215 a7ce10d Thanks @PuruVJ! - The dock's button no longer sticks on "loading…" when its code fails to download: a click asks for it again (#201).
  • #370 9272db9 Thanks @PuruVJ! - The devtools Page tab asks the dev server for a file's Vary header and for the pre-bundled packages again after a failed ask, instead of hiding those findings for the rest of the session (#363).
  • #409 44b3016 Thanks @PuruVJ! - The dock's save as a page goes (pagehide) no longer starts a request to the dev server, which WebKit reported as a page error.
  • #370 9272db9 Thanks @PuruVJ! - The devtools dock's server profile stops asking a busy profiler once its tab is left or the dock closes, follows the busy answer's Retry-After, and waits as long as the profiler's run page does (one shared wait) (#368).
  • #255 fc35032 Thanks @PuruVJ! - The dock's failed download is now logged as a message code with its page, devtools_dock_load_failed, like every other devtools message (#239).
  • #222 ff827ab Thanks @PuruVJ! - Finding what each garbage-collection pause allocated takes time in line with the profile.
  • #187 b46798f Thanks @PuruVJ! - Finding what ran at each garbage-collection pause takes time in line with the profile's pauses and timeline.
  • #187 b46798f Thanks @PuruVJ! - Finding a load's constant work takes time in line with the file's helpers: a helper that answers per call marks the helpers that call it, with no pass over every helper again.
  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.
  • #340 24c6592 Thanks @PuruVJ! - ogygia_profile no longer takes over someone else's live recording: the profiler's 409 says who holds the recorder and whether that holder is stale, and the tool resets only a stale one; a live one gets a "busy" message (#290).
  • #370 9272db9 Thanks @PuruVJ! - A profiler trap or always-on sampler whose window the inspector refused now says so: the dashboard shows what the trap really caught, stopped with the reason, and a window started later no longer shows a stopped trap as armed; the stop is said once as devtools_profiler_background_stopped (#361).
  • #370 9272db9 Thanks @PuruVJ! - A profiler mark() made before the page's beacon loads now waits for it however late it comes, instead of being dropped after 10 s; the beacon takes the waiting marks as it loads (#369).
  • #321 17e009a Thanks @PuruVJ! - A profiler that cannot start (an edge runtime without node:inspector, a profiler module that fails to load) now prints devtools_profiler_mount_failed once, instead of leaving every request unprofiled without a word.
  • #255 fc35032 Thanks @PuruVJ! - The server's first requests that arrive together now mount one profiler instead of one each, and a profiler that failed to load is tried again by the next request instead of staying off until a restart (#242).
  • #306 8b33368 Thanks @PuruVJ! - The profiler stops reading the copy of an outbound body it counts when the request that made the call ends, so a body that never ends is not read for good.
  • #306 8b33368 Thanks @PuruVJ! - After the hooks run again in dev, the new profiler sees every outbound fetch and http call, and http is wrapped once (#285).
  • #225 29d8553 Thanks @PuruVJ! - The profiler's recorder has one owner: a recording that a reset, a second stop or another profiler outlived no longer frees the next one's lock or switches off its spans, call stacks, network log or kept answers, a reset clears the background window too, and a window past its cap that fails leaves no unhandled rejection (#196).
  • #321 17e009a Thanks @PuruVJ! - The profiler's Redis store, given a URL, fails with node-redis's own connection error (a wrong password, an unreachable host) instead of "Cannot find package 'ioredis'".
  • #306 8b33368 Thanks @PuruVJ! - The profiler's run view lets go of its request, its queue's waits and its progress timer when it unmounts.
  • #323 7a53a56 Thanks @PuruVJ! - The profiler's sink posts go through an outbox, one at a time in order: rows read while a post is out are sent even when the profiler stands down meanwhile, a failed post is tried again before newer rows go (so a full buffer drops the oldest rows, not the newest), each post carries an idempotency-key, and a post never taken is said as devtools_profiler_sink_undelivered (#293).
  • #323 7a53a56 Thanks @PuruVJ! - A profiler sink that is down holds one post at a time: the rows read meanwhile wait in the capped buffer (which drops the oldest when full) and go in one post after it, instead of a post queued per tick without a bound (#376).
  • #321 17e009a Thanks @PuruVJ! - The profiler's SQLite store fails with the driver's own error when it cannot open its file, instead of saying it needs Node 22.5+ or better-sqlite3.
  • #305 e605070 Thanks @PuruVJ! - A durable profiler store whose init fails once is tried again by its next use instead of being dropped for the life of the process (#289).
  • #225 29d8553 Thanks @PuruVJ! - The profiler's Postgres and Redis stores connect and set up once: a setup that fails at the first write is tried again by the next one instead of breaking the store until a restart, and two first calls at once share one client (#197).
  • #370 9272db9 Thanks @PuruVJ! - A file the static host could not answer once (a 5xx, a timeout, a network blip) is weighed again by the next profiler report instead of being left out for the life of the process, and reports weighing a file at once share one fetch (#362).
  • #255 fc35032 Thanks @PuruVJ! - Stopping a session recording a second time no longer ends the recording started after it, nor takes back its console patch: a recording's stop runs once and touches only its own run (#240).
  • #184 1767110 Thanks @PuruVJ! - Packages declare sideEffects (only their .css files run on import), so bundlers drop the modules an app never uses.
  • #401 3ec7cfe Thanks @PuruVJ! - The profiler's SQLite store prunes a page's visits by walking its index past the kept ones, instead of rebuilding the whole kept set on every write.
  • #254 346c692 Thanks @PuruVJ! - The testing engine behind __ogygia_testing that fails to download is asked for again under a fresh URL by the next call, instead of every call failing until a reload (#241).
  • #187 b46798f Thanks @PuruVJ! - Finding the waits in a row of a line takes time in line with its calls: the calls that started together are found from one sort.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/federation

Patch Changes

  • #221 a5d28b0 Thanks @PuruVJ! - The deploy thaw is tried again when its first check fails (the store unreachable, or a peer that never took the notice), instead of being lost until the pages' TTL, and only the peers that missed it are sent it again, for a bounded number of tries (#198).
  • #221 a5d28b0 Thanks @PuruVJ! - After a thaw or a mutation, an old peer request ending no longer removes the newer one, so a read joins it instead of starting another upstream fetch (#210).
  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.
  • #333 865eb55 Thanks @PuruVJ! - A peer read that starts while a mutation to that peer is out is no longer kept once the mutation ends, so the poster's next read never gets the page from before their own post (#331).
  • #305 e605070 Thanks @PuruVJ! - A thaw drops only the peer documents it names: a fetch already out for another fragment is kept when it lands, and a caller whose fetch a thaw or a mutation made stale reads the page again instead of getting the old one (#294).
  • #323 7a53a56 Thanks @PuruVJ! - A publish's thaw notices go through an outbox handed to the request's waitUntil (the new app.keepAlive), so a serverless instance that freezes after the response no longer loses them and peers no longer keep stale pages until their TTL (#286).
  • #321 17e009a Thanks @PuruVJ! - A peer whose store fails to apply a thaw notice now answers an error, so the sender sends it again, instead of answering done and deduping the retry; app.notify rejects once every hook ran when one failed (#287).
  • #321 17e009a Thanks @PuruVJ! - A thaw notice that a peer never took, after every retry, now prints federation_thaw_notice_failed with the peers that missed it, instead of nothing.
  • #323 7a53a56 Thanks @PuruVJ! - A peer applies two copies of a thaw notice that arrive at once only once, and keeps at most the newest 10 000 notice ids instead of every id for ten minutes (#308).

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected], @ogygia/[email protected]

@ogygia/flags

Patch Changes

  • #256 25ab8c7 Thanks @PuruVJ! - An exposure sink that rejects no longer crashes the server: the failure is logged with a message code, and the sink's send is handed to the platform's waitUntil when it has one, so serverless instances no longer cut exposures off. The response never waits for the send (#243).
  • #323 7a53a56 Thanks @PuruVJ! - Exposure batches go through an outbox: a batch the sink failed is tried again before the batch read after it, instead of being dropped, batches still reach the sink one at a time in order, and a send started by the size or age cap is kept alive past the response too (#308).
  • #340 24c6592 Thanks @PuruVJ! - A flag source that throws before it awaits anything leaves the request resolved (every flag on its native rule), instead of priming for good (#374).
  • #340 24c6592 Thanks @PuruVJ! - A second prime_flags of a request while the first still waits on the source now waits for that answer, instead of returning at once and leaving reads on the native rule (#296).
  • #321 17e009a Thanks @PuruVJ! - A flag source that fails (a throwing source, an OFREP endpoint that errors or cannot be reached, an OpenFeature flag whose evaluation throws) now prints flags_source_failed once per failing streak, instead of falling back to the native rule without a word.
  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.
  • #184 1767110 Thanks @PuruVJ! - Packages declare sideEffects (only their .css files run on import), so bundlers drop the modules an app never uses.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/machine

Minor Changes

  • #430 34fcfc8 Thanks @PuruVJ! - A flights keyed by some fields of its input is made with flights.keyed<I>()({ key, run }); flights() with no options no longer makes one.
  • #329 c781aeb Thanks @PuruVJ! - flights({ retry }) takes a policy() (tries, waits between them, deadline) in place of retries, run by the same loop as the outbox's; a wait of 0 still tries again at once (#313).
  • #304 b4113bf Thanks @PuruVJ! - flights(): keyed async work that callers share (one flight per key, kept as long as keep says, max/weigh, drop, fresh revalidation, work stopped when its last caller leaves, and a failure never kept), in place of once(), which is a flights with one key.
  • #379 11c5f0d Thanks @PuruVJ! - mailbox() runs one owner's events one at a time, in the order they came: a handler's turn lasts until it waits or ends, and the owner closing drops the queue and ends every wait with STALE.
  • #429 0d22e3e Thanks @PuruVJ! - backoffAfter(), MACHINE_ERRORS and MACHINE_WARNINGS are exported from the package index; the ./internal/messages and ./internal/policy subpaths are gone.
  • #323 7a53a56 Thanks @PuruVJ! - outbox(): work after the response, kept alive (waitUntil), tried again by a named retry policy() (attempts, backoff, jitter, deadline), joined by id while out, sent inOrder when asked, done only once the receiver applied it, and said once with a code when given up (#308).
  • #188 9cd69de Thanks @PuruVJ! - New package: typed state machines (machine(), lock()) with checked moves, async work that cannot land late, cleanups that run exactly once, states that end on their own (after), and the definition as data (graph(), paths()).
  • #304 b4113bf Thanks @PuruVJ! - scope(): one owner for listeners, timers, observers, tasks, child scopes and cleanups, all let go together and exactly once on close(); each visit of a machine to a state is one (m.scope), in place of task() and onLeave().
  • #324 7b35e33 Thanks @PuruVJ! - stage(): a change built off to the side and switched in with one synchronous step, owned by a scope: closing the owner first gives back what the build made, and a commit that throws puts back what it changed (#309).
  • #327 79aa5b3 Thanks @PuruVJ! - world(): a stamp that only moves forward (stamp(), fresh(s), bump()), split by key with world.keyed() (bump(keys) puts out of date only what read those keys); flights({ world, touches }) drops what a move touches and makes an answer whose world moved on its way STALE for every caller that joined it (#311).
  • #379 11c5f0d Thanks @PuruVJ! - A wait gives STALE when anything it waited on is stale: scope.task, a mailbox's waits and ask, a flights run and stage().ready turn STALE or an array with STALE in it (a Promise.all with a stale part) into STALE, and Fresh<T> types the slots after the check (#378).
  • #328 cc047cc Thanks @PuruVJ! - flights.keyed<I>()({ key, ignore, run }): a flights over an object input is keyed by the fields key names, spelled one way, and run is typed to those fields only, so work that reads a field the key leaves out does not compile unless ignore gives the reason; peek and drop name keys by those fields, and a flights with no key is keyed by its whole input (#312).

Patch Changes

  • #360 64431ac Thanks @PuruVJ! - Over its max, a flights takes out the expired answers before any live one (#356).
  • #360 64431ac Thanks @PuruVJ! - A flights run that gets its own key before it returns joins its own flight and gets its answer, instead of STALE (#354).
  • #379 11c5f0d Thanks @PuruVJ! - A flights run that answers STALE no longer leaves its key flying: nothing is kept, and the next get runs again (#377).
  • #360 64431ac Thanks @PuruVJ! - A keyed flights no longer holds every symbol used as a key field forever: a symbol made at run time is freed once nothing else holds it, and a registered one is spelled by its name (#357).
  • #360 64431ac Thanks @PuruVJ! - A flights answer whose weigh is below 0 or not a number is not kept, so a bad weight can no longer lower the total and let max cap nothing (#355).
  • #360 64431ac Thanks @PuruVJ! - In a production build, a step naming a state that does not exist no longer makes wait() and paths() throw: they skip it, as moves already do (#358).
  • #360 64431ac Thanks @PuruVJ! - An outbox item whose id is not a string or a finite number (undefined, NaN) is not sent and is given up, instead of joining every other such item into one send (#350).
  • #360 64431ac Thanks @PuruVJ! - A retry that stops because its next wait would end past the deadline fails with a TimeoutError (the last error in its message), and a policy's wait is always a number, also for a count of 0 or a 0 backoff after very many tries (#359).
  • #360 64431ac Thanks @PuruVJ! - The wait between two tries of a send (an outbox's, a flights({ retry }) run's) keeps a Node process alive, so a process with nothing else to do no longer exits mid-retry and loses the send (#351).
  • #360 64431ac Thanks @PuruVJ! - An async disposable adopted by a scope that already closed is disposed at once and its failure is said (machine_dispose_failed), not left unhandled; the closed scope keeps nothing of it (#347).
  • #360 64431ac Thanks @PuruVJ! - A scope's { once: true } listener lets go of itself when it fires, as a fired timer does, so a long-lived scope no longer piles them up (#348).
  • #360 64431ac Thanks @PuruVJ! - A scope whose listener release throws on close() still lets go of the rest and runs every cleanup, then throws it (#346).
  • #360 64431ac Thanks @PuruVJ! - A scope task whose work closes its own scope before it returns (moves its machine on) settles with STALE instead of hanging (#349).
  • #360 64431ac Thanks @PuruVJ! - A stage whose commit closes its own owner (switching the page in moves the navigation on) keeps what it committed instead of discarding it mid-commit (#343).
  • #360 64431ac Thanks @PuruVJ! - A stage's discard step given after the commit is dropped instead of run at once over the live change (#344).
  • #360 64431ac Thanks @PuruVJ! - A stage that committed or was discarded leaves nothing on its owner, so a long-lived owner of many stages no longer grows (#345).
  • #360 64431ac Thanks @PuruVJ! - A world holds the flights made on it weakly: a flights nobody holds any more is collected and the world forgets it, instead of each one staying until disposed (#352).

@ogygia/postrender

Minor Changes

  • #177 d9e8e2f Thanks @PuruVJ! - Postrender's stores and edges are classes: new MemoryStore(), new ValkeyStore(client), new UpstashStore({ url, token }), new CloudflareKvStore(binding), new AkamaiEdge(cfg), new CloudFrontEdge(cfg) and new CloudflareEdge(cfg) replace memoryStore(), valkey(), upstash(), cloudflareKv(), akamai(), cloudfront() / awsCloudfront() and cloudflare() (same arguments).

Patch Changes

  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.
  • #217 739b763 Thanks @PuruVJ! - A render that a publish (or any invalidation) lands during is served but not stored, so the page as it was before the publish is no longer kept (#194).
  • #217 739b763 Thanks @PuruVJ! - A stampede flight settles however the request that opened it ends, so a plugin answering that request no longer leaves every later cold request for the page waiting (#194).
  • #305 e605070 Thanks @PuruVJ! - Requests waiting on a page render that a publish overtook are let go at once and share one new render, which is stored, instead of waiting for the old render to end and then each rendering the page alone.
  • #217 739b763 Thanks @PuruVJ! - A response refused after its body was read is sent rebuilt from that body, not as the spent original.
  • #333 865eb55 Thanks @PuruVJ! - A page whose render was stored just after a publish evicted it is no longer served while it is taken back out, and that evict is now tried again when the store fails it (#332).
  • #321 17e009a Thanks @PuruVJ! - A store that is down now prints postrender_store_failed once per failing streak, instead of every render quietly missing and nothing being kept.
  • #370 9272db9 Thanks @PuruVJ! - A store outage is said once (postrender_store_failed) even while calls overlap: a slow store call that started before the outage and answers after it no longer ends the failing streak (#364).
  • #323 7a53a56 Thanks @PuruVJ! - A successful form post's response now waits for its page's stored copy to be evicted (an outbox send, tried again within 2 s), so the redirect's GET right after renders fresh instead of getting the page from before the post (#283).
  • #321 17e009a Thanks @PuruVJ! - A peer whose store fails to apply a thaw notice now answers an error, so the sender sends it again, instead of answering done and deduping the retry; app.notify rejects once every hook ran when one failed (#287).

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/router

Patch Changes

  • #226 019b77e Thanks @PuruVJ! - A streamed page drains its late regions until nothing new registers: a late region that resolves to a region holding another promise region streams that one down the same response too, instead of leaving an empty slot nothing fills (#199).
  • #321 17e009a Thanks @PuruVJ! - api() rejects when a successful answer's JSON body does not parse, instead of resolving undefined.
  • #306 8b33368 Thanks @PuruVJ! - A streamed page whose visitor leaves stops: its generator and every late-region source end, and nothing more renders (#284).
  • #321 17e009a Thanks @PuruVJ! - A streamed page's source whose cleanup throws as the visitor leaves now prints request_cleanup_failed with the error, instead of nothing.
  • #184 1767110 Thanks @PuruVJ! - Packages declare sideEffects (only their .css files run on import), so bundlers drop the modules an app never uses.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/testing

Minor Changes

  • #177 d9e8e2f Thanks @PuruVJ! - The testing primitives over any driver are a class: new Ogygia(driver, options) replaces createOgygia(driver, options), and a driver's own get, post and wait are now called as its methods.

Patch Changes

  • #175 a99a49a Thanks @PuruVJ! - Every ogygia error and warning has a code and a page at ogygiajs.dev/e/<code>: dev prints the wording and the link, a production build prints the link alone ([ogygia] https://ogygiajs.dev/e/<code>?<args>), so the wording stays out of the client and server bundles.
  • #306 8b33368 Thanks @PuruVJ! - record() whose actions fail throws their error even when an undo (the navigation blocker's or the server window's stop) fails after them, instead of the undo's (#372).
  • #184 1767110 Thanks @PuruVJ! - Packages declare sideEffects (only their .css files run on import), so bundlers drop the modules an app never uses.
  • #306 8b33368 Thanks @PuruVJ! - record() that fails early stops its server window, untags the requests and lifts the navigation blocker, so the next record() starts at once (#291).

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

ogygia

Patch Changes

  • #184 1767110 Thanks @PuruVJ! - Packages declare sideEffects (only their .css files run on import), so bundlers drop the modules an app never uses.

Updated dependencies: @ogygia/[email protected]

0.12.0 minor

@ogygia/cli

Updated dependencies: @ogygia/[email protected]

@ogygia/content

Updated dependencies: @ogygia/[email protected]

@ogygia/core

Minor Changes

  • #157 a217b91 Thanks @PuruVJ! - An island loaded without a src (a region the browser renders, a held or live region from a remote function or a stream, a hole rebuilt after a failed hydration, a portable snippet with no island graph) loads its content-hashed file: the build writes its locations into one content-hashed map, which the runtime's script names and the first such load reads.
  • #157 a217b91 Thanks @PuruVJ! - A client-side navigation into a page of another build (a deploy since the page was loaded) loads that page as a document instead of swapping it into the running runtime, so two builds' islands, runtimes and Svelte never meet on one page.
  • #157 a217b91 Thanks @PuruVJ! - Every file a build writes under <appDir>/immutable/ is now named by its content, so a host's year-long rule for that directory can never serve another build's code: the stable-name files (og-region.<id>.js, og-runtime.<id>.js), whose content changed every build, are gone. An island's entry is now a name, not a file. The one file at a fixed name is <appDir>/ogygia.json, beside Kit's version.json: the current build's locations, which a page that outlived its build reads to wake its islands on the current code. Serve it as you serve version.json (revalidated); a federated app also needs the same Access-Control-Allow-Origin on it as on its immutable/ files. A page cached from before content hashing, which loaded the runtime by its stable name, no longer recovers.

Patch Changes

  • #151 64f6046 Thanks @PuruVJ! - A client-side navigation no longer removes nodes the page's own code put in <body> (a popover portal's anchor, a toast host), so a popover in an island the navigation keeps still opens afterwards; only the server's body children and attributes are replaced.
  • 75a5cdd Thanks @PuruVJ! - handle({ secret }) signs region capabilities with a secret the app passes (e.g. from $env/static/private), for a server that never sees OGYGIA_SECRET at runtime; it signs exactly as the same OGYGIA_SECRET does, and every copy of ogygia in the process uses it.

@ogygia/devtools

Patch Changes

  • #157 a217b91 Thanks @PuruVJ! - The profiler's standalone report inlines each island by the file its src names, so its islands wake offline.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/federation

Patch Changes

  • #157 a217b91 Thanks @PuruVJ! - A remote fragment's island graph (its chunk preloads and portable snippets' locations) is pinned to the remote's origin like its attributes, so the shell no longer resolves it against its own files.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/flags

Updated dependencies: @ogygia/[email protected]

@ogygia/postrender

Updated dependencies: @ogygia/[email protected]

@ogygia/router

Updated dependencies: @ogygia/[email protected]

@ogygia/testing

Updated dependencies: @ogygia/[email protected]

ogygia

Updated dependencies: @ogygia/[email protected]

0.11.0 minor

@ogygia/cli

Patch Changes

  • #83 e229a9a Thanks @PuruVJ! - ogygia perf --margin 0 is accepted, and ogygia ai and the MCP tools describe the current devtools setup.
  • #83 e229a9a Thanks @PuruVJ! - ogygia site init generates a working search page (it called an unimported site).

Updated dependencies: @ogygia/[email protected]

@ogygia/content

Updated dependencies: @ogygia/[email protected]

@ogygia/core

Patch Changes

  • #83 e229a9a Thanks @PuruVJ! - Remote deferred holes now share local holes' lifetime and renewal: they were signed for a fixed 5 minutes and answered 410 after, so a stored or cached page older than that showed dead remote regions. Plugins sign with core's rule through app.capability.

@ogygia/devtools

Patch Changes

  • #88 24c2845 Thanks @PuruVJ! - The profiler UI's display font is named Geist Serif, matching the docs site.
  • #83 e229a9a Thanks @PuruVJ! - The profiler's sqlite, redis and postgres stores keep a whole recording window (2000 requests per key); they kept the newest 20.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/federation

Patch Changes

  • #83 e229a9a Thanks @PuruVJ! - Remote deferred holes now share local holes' lifetime and renewal: they were signed for a fixed 5 minutes and answered 410 after, so a stored or cached page older than that showed dead remote regions. Plugins sign with core's rule through app.capability.
  • #83 e229a9a Thanks @PuruVJ! - A peer-cache refresh that answers an error no longer replaces the good document it was refreshing.

Updated dependencies: @ogygia/[email protected], @ogygia/[email protected]

@ogygia/flags

Patch Changes

  • #83 e229a9a Thanks @PuruVJ! - A flag source receives the visitor by the same rule a flag read uses, so openfeature() and ofrep() send a targetingKey by default.

Updated dependencies: @ogygia/[email protected]

@ogygia/postrender

Patch Changes

  • #83 e229a9a Thanks @PuruVJ! - The deploy-thaw build marker no longer lapses after a day: it is stored apart from the page TTL and rewritten every boot.
  • #83 e229a9a Thanks @PuruVJ! - A stored page older than the policy ttl is a miss on read, so a store without its own expiry still ages pages out.
  • #83 e229a9a Thanks @PuruVJ! - A store that fails during invalidation is reported as a store failure, and the edges still purge (a throwing evictByTag stopped the whole invalidation and was blamed on the edge).

Updated dependencies: @ogygia/[email protected]

@ogygia/router

Updated dependencies: @ogygia/[email protected]

@ogygia/testing

Updated dependencies: @ogygia/[email protected]

ogygia

Minor Changes

  • #64 3bff299 Thanks @PuruVJ! - Breaking: features move into plugin packages, each a Vite plugin listed beside ogygia() and before sveltekit() (plugins: [ogygia(), flags(), sveltekit()]; out of order, or without ogygia(), the build says so): flags (@ogygia/flags, was ogygia/flag), frozen pages (@ogygia/freeze, was ogygia/freeze and ogygia({ freeze })), fragment federation (@ogygia/federation, was ogygia/federation; mount moved from ogygia/router) and content (@ogygia/content, was ogygia/content* and ogygia({ content }); ogygia/content/server is @ogygia/content/server, and the svelte config's ogygia.preprocess() / ogygia.extensions() are preprocess() / extensions() from @ogygia/content/vite), each with its factory from <package>/vite. The devtools dock and the profiler move to @ogygia/devtools: ogygia({ devtools, profiler }) becomes devtools({ dock, profiler }), ogygia/profiler* imports become @ogygia/devtools/profiler* (@ogygia/devtools/profiler, /profiler/client, /profiler/storage, /profiler/storage/sqlite|redis|postgres), and the event layer ogygia/devtools exported is ogygia/events (with ogygia/insights beside it). ogygia/testing is @ogygia/testing. The server router is @ogygia/router (was ogygia/router; list router() from @ogygia/router/vite so its pages' CSS links — dev warns when a page renders without it), and its typed fetch client ogygia/router/client is @ogygia/router/api (POST(schema, …) endpoints now type-check, and $infer carries their verbs). An endpoint handler's c.params is now typed strictly by its table key, at the top level, inside layout() tables and behind when(), and its return type reaches api(); GET<'/p'>(fn) is gone (a handler written outside its table checks with satisfies Handler<'/p'>). A plugin is made with definePlugin({ name, server, client, options, build }) from ogygia/vite, which returns a Vite plugin; its server and browser halves are named by location (server: import.meta.resolve('./server.ts')) instead of found by package export name. The command line lives in @ogygia/cli (npx ogygia … keeps working: ogygia's bin runs it). ogygia is now a thin package over @ogygia/core; plugins depend on @ogygia/core. ogygia()'s region options sit at its root: regions.visible, regions.presets, regions.ttl and regions.maxPageAge are visible, presets, ttl and maxPageAge. @ogygia/tsdown-plugin-svelte turns a component's relative ./x.ts imports into ./x.js in the shipped component and its types, so sources can import with real extensions. Fixed: a Kit-hydrated (csr = true) page could link the stylesheets of every region in the app, depending on how the bundler grouped the server build (newer rolldown, under vite 8.3, did).

Patch Changes

  • #77 fd47ef6 Thanks @PuruVJ! - Breaking (plugin API): a plugin's per-route setting is now a key inside Kit's own config page option — b.routeConfig(key) / ctx.routeConfig(key) replace b.routeOption(name) / ctx.routeOption(name), nothing is stripped from route files any more, and the server router's page() / layout() / routes() take config instead of freeze.
  • #80 5164260 Thanks @PuruVJ! - Server router: a page's load and actions (bare, or through load() / action()) read c.params typed by their table key, as endpoints do, and a params or search schema now types c.params / c.search with its output.
  • #74 1d97ecc Thanks @PuruVJ! - Types: import.meta.og.source is declared, instrument's attrs callback may name fewer arguments than the function takes, and loose anys in public types are now unknown (OgygiaDriver.evaluate, mount()'s page data, the router's component type); the profiler timeline's blocks are buttons.

Updated dependencies: @ogygia/[email protected]

0.10.0 minor

ogygia

Minor Changes

  • #33 6f48146 Thanks @PuruVJ! - Breaking: a render: 'deferred' hole's timing is the new fetch key (with { render: 'deferred', fetch: 'visible' }); pairing it with wake fails the build with the line to write, and wake now only means when an island's JS runs.
0.9.0 minor

ogygia

Minor Changes

  • #51 c0a7793 Thanks @PuruVJ! - Raised minimums to the versions ogygia is tested on: svelte ^5.56.8, playwright and @playwright/test ^1.62.0 (for ogygia/testing), and the magic-string dependency ^1.4.5.
  • #51 c0a7793 Thanks @PuruVJ! - SvelteKit 3 support, alongside SvelteKit 2: builds no longer fail on refreshAll (#37), islands on csr=true pages get Kit's page, the $page store no longer uses the removed $app/stores, custom-type remote arguments encode on csr=false pages, and a remote form in an island on a csr=true page no longer reloads the page.
  • #51 c0a7793 Thanks @PuruVJ! - The vite peer is ^8.2.4: Vite 8, from the first version past its published advisories. SvelteKit 2 and SvelteKit 3 both run on it.

Patch Changes

  • #51 c0a7793 Thanks @PuruVJ! - devalue is required at ^5.9.3, the first version with the fixes for its published advisories.
0.0.0 minor

@ogygia/cli

@ogygia/content

@ogygia/core

@ogygia/devtools

@ogygia/federation

@ogygia/flags

@ogygia/postrender

@ogygia/router

@ogygia/testing

@ogygia/tsdown-plugin-svelte

ogygia

0.8.0 minor

ogygia

Breaking changes

  • Flags and experiments collapse onto one primitive, flag(); experiment, layer, allowOverrides, onExposure, batchExposures, .bucket, .on and routes({ experiments }) are removed. Declare with a shape (flag('x'), flag('x', 10), flag('x', fn), flag('x', { control: 80, bold: 20 })), read by calling it (checkout(c)), branch with flag.pick(map) / flag.pick(c, map), read payloads with flag.value(c), configure once with decide({ source, overrides, exposure, batch }), and pre-decide in a route-only shell with routes({ flags: [...] }).
  • kitMount is now mount.kit.
  • ogygia/rewrite is now ogygia/markup (no alias; change the import).
  • transformHole is removed; ogygia.handle({ transform }) covers region answers and documents alike.
  • regionTtl is now regions: { ttl }.
  • regions.preload is removed; each island's whole graph preloads at its wake.

Added

  • One load scheduler for every download ogygia starts: island code and background work wait for the page's fetchpriority="high" resources (or the first input, or a 2.5 s cap), so islands no longer compete with the hero image.
  • Holes and regions a remote query / command renders see their page's page.url, route, params and server-load page.data (looked up through Kit, only when read), so a server-rendered header can refresh in place ("A lake that can refresh").
  • One DOM ownership model for every ogygia DOM writer — who writes inside an element, its attributes, its children's order and their existence — so a web component's own DOM survives repairs, hole answers and navigations.
  • SvelteKit 3 support, on the APIs late Kit 2 and Kit 3 share.
  • The SSR profiler, ogygia({ profiler }): CPU, network, heap and the browser half of a visit, with a 0–100 page score and on-demand mode (profiler: { onDemand: true }).
  • Devtools, ogygia({ devtools }): an island-graph dock with Page, Hydration, Record, Bytes and Profiler tabs, every vital explained down to the island and line responsible.
  • ogygia/testing: the devtools and the profiler as test primitives.
  • ogygia/router: Kit's routing, programmatic and fully typed; document() renders a complete page from a handle.
  • ogygia/internal/compiler: the compiler as a standalone, bundler-agnostic engine.
  • Frozen pages, ogygia({ freeze: true }): a page renders once on write and is served as stored bytes until a publish thaws it (with edge stitching for per-visitor holes).
  • Fragment federation: one federate(), a remote fragment is a region, and a publish thaws it everywhere.
  • The Ref hub: one identity layer for live classes, stores, functions, snippets and held regions that cross an island boundary.
  • Transportable values: everything devalue rejects now crosses through one serialization seam; unbridgeable values fail at discovery, not in production.
  • A navigation moves regions instead of resetting them (state-delta reconciler); opt-in server-delta navigation renders only what changed (router: { serverDelta: true }).
  • ogygia.handle({ transform }): server transforms that may reshape Svelte-owned markup, restored in the browser before hydration.
  • ogygia/markup and scanRegions(html): the region round trip for an app that runs a third-party SSR pass over the page.
  • On-demand server islands: render: 'deferred' + wake: 'interaction' fetch nothing until the visitor shows intent.
  • prefetch warms a deferred hole's HTML before it swaps.
  • keepFallback(): a server island's answer that keeps the page's fallback.
  • requestEvent(): Kit's RequestEvent inside a server island.
  • isOgygiaPage(): tells shared code which world it runs in.
  • A lake survives Kit hydration on a csr = true page.
  • ogygia({ barrels }): barrel imports become leaf imports at transform time.
  • Zero-config library islands: a dependency declares its island roots in its own package.json.
  • The app's hooks.client.ts init runs on csr = false pages.
  • Server islands ship their component tree's CSS; region CSS obeys Kit's inlineStyleThreshold.
  • Lazy islands download at background priority, so they never fight the first paint.
  • Accessible SPA navigation (focus, announcements) and a reason on every page fetch.
  • when(): flags gate routes, and pick chooses infrastructure; OpenFeature interop (ogygia/flag/openfeature).
  • The build warns when island code reads a context nothing bridges.
  • A recovered or healed island says why, and the discard warning names what Svelte threw.
  • The ogygia MCP server grows to eleven tools, and the bundled AI skill is rewritten.

Changed

  • A <Region of={promise}> waits where the render can: awaited in the server render under Svelte's async mode, streamed as a late region on the server router, or resolved after hydration otherwise.
  • Islands inside a region's HTML always wake on the runtime, csr = true pages included.
  • The page seed ships only when an island reads it, and only the page.data keys the page's islands read.
  • The remote seed ships only for remotes an island on the page can call.
  • Island props reference the page seed instead of copying it, and ride at the end of the body with the module-preload hints.
  • Island hints are fetchpriority="low"; island wakes start after the page has painted.
  • The boot never loads Svelte and is four files fetched in one round trip; the runtime goes right before the page's own JavaScript.
  • Duplicate stylesheet links are dropped.
  • HTML is byte-identical across renders (window-aligned capability expiry, per-request slot ids).
  • The server request path is one request store, one document pass and one walk, and stops allocating per item.
  • A slow or hung page can no longer wedge the profiler or the site, and recordings coordinate across processes.
  • The Vite plugin is a compiler, and region rendering is region-granular.
  • rolldown is no longer a dependency: the compiler parses and bundles through the app's Vite, keeping its 16 MB native binary out of production installs.

Fixed

  • Island entries on pages below the root no longer 404 in builds (Kit's asset() form comes from the build's Kit version).
  • A region moved by another script keeps its in-flight fetch and its woken island.
  • Restored shadow roots keep their sheets across a router navigation.
  • A preloaded load hole stays out of the batch instead of downloading twice.
  • Region URLs are pinned to their page, there is one region per hole across path depths, and a navigated island arrives whole in one swap.
  • A registry's plain imports keep their CSS on csr = false pages; an island inside an island links its own CSS; held regions keep inline sheets.
  • A kept island adopts its server DOM instead of discarding it on every load.
  • An island with children has the same fingerprint and bytes on every render.
  • An island edited while it slept hydrates from its own server markup instead of re-rendering.
  • A hole answer that is not the region's (a redirect, a whole document) is refused.
  • A deferred hole survives Kit rebuilding a client-on document, and outlives a CDN-cached document.
  • A hole's answer morphs over its fallback on every schedule, keeping a web component's attributes and state, unfocused edited fields, and focus visible to assistive tech.
  • Region and hole HTML is parsed aware of declarative shadow DOM.
  • A lake with no island host survives Kit giving up on the document, and a lake's island with props hydrates on a csr = true page.
  • An error page under a csr = false page renders as the Kit-hydrated document it is.
  • A csr = true route that imports a block registry no longer links every block's CSS and chunk, and a csr = false page links only the island CSS it renders.
  • Island entries and the runtime are content-hashed, so a deploy never runs the last build's island code.
  • The runtime's hash module is no longer named fingerprint.js, which ad blockers blocked.
  • A visible island's code waits for the viewport, and a scroll-woken island no longer waits on the scheduler.
  • An island of ours on a Kit document waits until Kit has applied its page; requestEvent() answers under Kit's own render too.
  • goto() from an island on a Kit-booted page navigates through Kit, and a script module reached through an app alias gets the island's page.
  • The router keeps a destination page's inline <style> sheets, follows data-sveltekit-reload's grammar, navigates links inside web components in place, corrects the address bar after a server redirect, and keeps the live <body> on a fallback swap.
  • An unkeyed wrapper around an island survives a sibling inserted above it, and lakes restore by DOM position.
  • A layout's csr world derives from the pages it serves, and a csr = true page under a csr = false layout keeps the layout's chrome.
  • kit.files.routes, kit.outDir, appDir and paths.base are honoured everywhere, and route walkers follow symlinks.
  • Island chunks are emitted only for islands the server bundle can reach.
  • Region fingerprints no longer cost 20 ms per island; the profiler is free when it is not profiling.
  • $app/stores $page works inside every ogygia render root, and a server island's page.url is the referring page.
  • On a csr = true page a server island keeps its server-minted endpoint through Kit's hydration, and server-island HTML is absolutized for nested paths.
  • wake: 'interaction' replays the waking click on the next frame, into the deepest target.
  • A wake: 'none' lake in a +layout.svelte ships its scoped CSS, a lake's or island's scoped CSS keeps its Svelte hash, and the fouc-css fallback ships valid CSS.
  • An island inside a server island's ogygiaFallback stays a region, and keepFallback() from an inline-rendered server island fails with the reason.
  • A relative with { wake } import inside a {#snippet}, a forwarded snippet opening with {@const}, and a raw <style lang="scss"> no longer break or silently skip a component.
  • A portable snippet carries its host's style scope, a branded snippet runs in place, and store auto-subscriptions in crossing snippets hoist as snapshots.
  • An island's <svelte:head> survives a head other scripts edited, and an island's page seed follows dynamic imports.
  • The compiler registers only wired transportable classes, and a byte-order-marked source is edited at the right offsets.
  • Freeze: a root prefix clears shared stores, and tag sets keep their longest lifetime.
  • Dev: CSS authored in holes, regions, islands and frozen snippets applies on first load.
  • Dev: every island is a crawl root for Vite's dep scanner, so nothing is discovered late on a cold server.
  • Dev: an installed ogygia runs as one copy in the browser.
  • Dev: a dep re-optimization no longer strands the islands on an open tab.
  • Dev: server islands keep working after an HMR edit of their host file.
  • Dev: toggling a route's csr export refreshes the server's route set.
  • Dev: the nested-island warning no longer crashes SSR.
  • Devtools: the dock is styled in a build and under a CSP that refuses inline styles; a wait for a failing first-screen island is not blamed on ogygia.
  • Profiler: a page that leaks never takes the server down.
  • Profiler: page profiling survives serverless timeouts, and network capture is reliable.
  • Profiler: its renders ask as a page navigation, so streamed pages profile as visitors get them.
  • Profiler: login survives a large cookie jar, and the dev server runs one live profiler per process.
  • Profiler: beacon visits over 64 KB are sent, one beacon per Kit page.
  • Profiler: JSON and .cpuprofile links download, a .ogp moved between machines imports, and the UI ships TS-free templates.

Internal

  • Full TypeScript strict at the library level, with svelte-check in the library check and zero explicit any.
  • oxfmt is the workspace formatter.
  • The per-host csr context cascade is one per-document signal (documentIsCsrTrue()).
  • The on-demand-hole browser tests no longer depend on where the runner's pointer rests, and the weekly upstream watcher gained a baseline leg.
0.7.0 minor

ogygia

Breaking changes

  • Unified cross-island context: <Context of={ctx} value={v}> and ctx.get() are replaced by Svelte's own setContext / getContext, bridged across islands.

Added

  • $page.data (with form, error, status) works inside islands on csr = false pages, streamed load promises included.
  • Drop-in setContext: adopt cross-island context with an import swap.
  • A .ts registry with { wake } binding is mountable.
  • import.meta.og.asRegion(Comp, options): the barrel escape hatch for regions.

Fixed

  • A csr = false subtree under a csr = true ancestor layout islands correctly, also next to a commented-out csr export.
  • ogygia's own injected imports resolve to ogygia's own files, and its components compile under SSR when the app externalizes it.
  • A plain function passed as an island prop errors as a function, not as a snippet.
  • The runtime chunk URL busts on the feature set, not only on source.
  • region() recognizes a wake: attach binding, and keep splits the region dedupe key.
  • Two lakes of the same component restore to the right boxes.
  • A held-region import that is only text (a comment, a template literal) is left alone.

Internal

  • One shared region-option parser, one emitter per region kind, and side-channel emits centralized.
0.6.0 minor

ogygia

Breaking changes

  • Transportable codecs are declared with the wire macro (static [ogygia.wire] = …).
  • The Vite peer is ^7 || ^8 (5 and 6 dropped).
  • Island facades are emitted as og-region.<hash>.js.
  • continuity.speculate is removed; SPA mode never emits speculation rules.

Added

  • The config surface: one grammar per ogygia() subsystem.
  • The site layer, site() in ogygia/content, with shells and bricks (Frame, DocsShell).
  • The import.meta.og.* compile-macro family.
  • Region snippets, region.snippet(): a snippet that crosses the island boundary.
  • Awaitable regions: await region(Component, props) bakes the SSR HTML into the ticket.
  • preference() / preference.switch(): site-wide, no-flash visitor preferences.
  • A Markdown authoring dialect with VitePress-compatible containers.
  • MPA-mode native speculation rules with router: false.
  • Dev guards: mutating a captured host snapshot inside an island warns.
  • ogygia/profiler: a drop-in SSR profiler.

Changed

  • Schema layers merge instead of chaining.
  • Preloading is render-gated end to end.
  • svelte/server and the codec graph no longer reach the client.
  • Router prefetch also warms the incoming page's island modules.

Fixed

  • A directly used <Region> on a csr = true page renders inline in the Kit tree.
  • An island reading $app/stores / $app/state no longer bundles Kit's real client store.
  • Dev soft-CSS HMR is scoped to the page's own sub-app.
  • csr = false apps no longer need a token csr = true route.
  • folder() collections no longer come up empty on the dev server.
  • Childless islands no longer serialize a phantom children slot.
  • Nested islands no longer compile in O(2^depth).
  • import.meta.og.code() in a .svelte host is no longer discarded.
  • An interrupted navigation no longer logs an unhandled "Transition was skipped" rejection.
  • ogygia/content/slot resolves.
  • Shell reactivity and accessibility fixes (version switcher, element overrides, tab groups).

Security

  • A dev-server path traversal in the FOUC CSS virtual is closed.
  • The region batch endpoint rejects oversized bodies (413) before reading them.
  • A full audit of the signed-capability pipeline.
0.5.0 minor

ogygia

Breaking changes

  • The SPA router is global; the <Router/> component is removed.
  • Content render() and renderHtml() are replaced by the entry's body (a region) with get() + <Region>.
  • OgygiaBoundary is now Boundary.
  • ogygiaHandle is now handle, exported from ogygia/server (ogygia/hooks stays as an alias).
  • The default island endpoint is /🏝️.
  • Presets speak the render / wake grammar.
  • @ogygia/content is merged into ogygia.

Added

  • Live regions: LiveView over query.live.
  • Streaming server islands (opt-in, ogygia({ stream: true })).
  • Regions: region(Component, props), a server-chosen renderable placed like data, with dual-face regions and ogygia.transport.
  • content is part of ogygia (ogygia/content).
  • One config surface in ogygia({ … }), and a namespace API (import * as ogygia from 'ogygia').
  • npx ogygia init wires a SvelteKit project in one step.
  • Async regions: <Region of={promise}> owns the full wait.
  • blocks.resolve(tree, registry) resolves a data tree to regions.
  • Per-hole browser cache, maxAge.
  • ogygia.script(): an inline <script> from one plain object.
  • The Fallback<P> type for a deferred island's fallback slot.
  • ogygia/internal/compiler: the pure transform engine.
  • Zero-file all-csr = false apps.

Changed

  • Deferred holes are dynamic by default (Cache-Control: no-store).
  • ogygia.preprocess() is synchronous.
  • Content sources are trimmed; the mdsvex source is markdown().
  • The island endpoint is matched independently of paths.base.

Fixed

  • Router back/forward swaps the page, and prefetched pages are used on click.
  • The $app/state page snapshot reaches islands, and updates after SPA navigation in dev.
  • Lakes survive client hydration.
  • Content-page islands build, and a standalone client build keeps SSR island registrations.
  • Deferred regions hydrate on csr = true pages.
  • Phantom modulepreload chunks are dropped.
  • Island hydration adopts SSR roots in place (no "hero bounce" reflow).
  • No FOUC on the first navigation after a deploy.
  • Held-region CSS styles the page, in dev and on serverless adapters.
  • A live command refresh with only refresh keys works.
  • The morph keeps form state with a keyed diff.
0.4.0 minor

ogygia

Breaking changes

  • Portable island bindings: a marked import rewrites the binding itself to an island wrapper, so <A />, dynamic components and {#each} lists of components work; host children on island call sites are a build error (except ogygiaFallback).

Changed

  • Dedupe is identity-based: the same component and options make one wrapper and one client entry, however many call sites.
  • csr = false client hosts no longer pull island wrappers into Kit's client graph.
  • The plugin uses build.rolldownOptions on Vite 8.
  • Props are real Svelte props into the wrapper, and ogygiaFallback is a normal snippet prop.
  • Lakes are portable wrappers too.

Removed

  • The static-tag-only requirement, and the tag-hoist virtual modules.
0.3.0 minor

ogygia

Changed

  • A hydrate island's entry is an importable module URL in production, and the runtime no longer embeds an app-wide regions map.
  • Each hydrate island is emitted at a deterministic ogygia-island.<id>.js, and SSR renders the same virtual island module the client hydrates.

Fixed

  • Dev: asset() no longer rewrites Vite /@id/… URLs, and props stay the region's immediate sibling.
  • SSR and the client share one virtual tree, ending hydration mismatches on captured props.