skipLibCheck exempts .d.ts contents from being checked, so assertions
written inside wrnexus.generated.d.ts were never evaluated by tsc.
Emit them into wrnexus.generated.api-checks.ts instead, referencing
the WRNexusGenerated namespace's helper types (which stay in the
.d.ts). Track the new generated file in check:generated-types.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Reuse tokenizer.readBalancedBraces string/comment skipping (extracted as
skipLiteralOrComment) for both section detection and slicing, instead of a
second hand-rolled brace counter. Fixes truncation on braces inside strings
and false-positive sectioned detection from keywords inside comments/strings.
Also switch api-sections.ts errors from plain Error to LexError so parser.ts
upgrades them to ParseError with an offset.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Strips TypeScript from client function bodies when emitting browser
modules, so `wrnexus build` no longer fails on an annotated local.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`wrnexus build` failed on any client function whose body used TypeScript:
const requestBody: Record<string, unknown> = {}
error: Expected ";" but found ":"
Codegen copies a client function's body into the browser module verbatim.
It removes the types from the function's *signature*, which is what made
this easy to miss -- the emitted module looked transpiled, and only bodies
carried types through. The artifact is written as .mjs and read back as
plain JavaScript, so the failure surfaced as a syntax error in generated
code rather than at the .wrn line responsible.
Browser modules are now transpiled before they are written, at all three
sites that emit one (the production build and both dev-server paths).
Reproduced end to end: a page with an annotated body failed the build with
the reported errors, and after the fix builds, ships valid minified JS, and
runs -- the handler sets its state correctly in a browser.
Note: the same body is also embedded as a string for the CSP-safe fallback
interpreter, which still receives it untranspiled. The compiled module
shadows the fallback, so this is only reachable in the window before that
module loads. Left alone here because stripping it lives in codegen, which
also runs under Node in the editor bundle where the Bun transpiler is
unavailable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Client control blocks and loops, the dev-server rebuild recycle, the
editor's tag and completion handling, and the island mount/HMR fixes.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two faults found by driving the island demo in a real browser. Both were
silent: the markup, every asset, and all 48 island tests were correct
either way.
An island renders nothing until it mounts, so its placeholder is
zero-height, and IntersectionObserver does not treat a zero-area target
consistently -- client:visible islands mounted on one load and not the
next. Visibility for those is now decided from the element's own rect,
driven by scroll and resize; a placeholder with real size still uses the
observer. The strategy had no test at all, which is why this shipped.
After an island source edit the browser kept running the old code. The
rebuild worked and the file was refetched, but the loader imports a URL
that does not change, and the browser caches modules by URL. Remounts now
carry a generation the dev loader folds into the request.
Verified in the browser: mounts with start={3} as a number, clicks reach
React (3 -> 5), and an edit to Counter.tsx now shows the new text and
stays interactive.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The dev server got slower the longer it ran. Measured on the example app:
30 .wrn edits grew RSS from 117 MB to 137 MB and never gave it back, while
30 CSS edits cost nothing -- so the leak is exactly one retained module
identity per rebuild, not caches or file handles.
That is inherent to reloading a module in-process. Bun caches modules by
path, so a rebuild has to be given a new identity to be picked up at all,
and Bun has no API to unload the old one. At roughly 0.66 MB a rebuild, a
long editing session is several hundred megabytes of garbage that cannot
be collected.
The process now recycles itself past a rebuild threshold, exiting with the
RESTART_EXIT_CODE the CLI supervisor already respawns on; browsers
reconnect because the HMR client already retries. It waits for a quiet
period first so a live request is never cut off, and the threshold (300
rebuilds, about 200 MB) sits well above a normal session. Set
WRNEXUS_DEV_RECYCLE_AFTER to tune it, or 0 to switch it off.
Also bounds browserArtifactPaths and islandArtifactPaths, which are keyed
by content hash and so gained an entry per rebuild that was never read
again. Small next to the module leak, but unbounded is unbounded.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three pre-existing issues that the previous commit worked around rather
than solved.
Test global pollution. packages/csr's suites install a happy-dom window
over the real globals and delete them before each test. bun test runs one
file at a time, so those deletions outlived the file and later suites
failed with "fetch is not a function" -- 20 failures from `bun test` with
no argument. They now restore what they captured. The editor's Node tests
shim the vscode host by patching Module._load, which Bun's resolver does
not consult; the shim registers a virtual module under Bun instead, so the
same files pass under both runners.
Multi-cursor tag auto-close. The handler now closes the tag at every
caret. Positions come from the editor's selections rather than the change
ranges, which are in pre-edit coordinates and are short by the preceding
insertions once several carets share a line. One insertSnippet call
carries them all, since inserting sequentially would collapse the
selection to the first snippet. Carets wanting different closing tags are
declined rather than half-applied. Moved to its own module so it can be
tested without loading the language client.
Runtime size. Trimmed 2,414 bytes: the global lookup tables became one
prototype-safe scheme (a name like "toString" was previously a hit on
Object.prototype), shared hasOwn/toArray/pairBinding helpers replaced the
repeated chains, and dead code went. That was everything available without
dropping or deferring a feature -- 49,000 was not reachable, so the budget
is now 50,500, set just above the real figure so future growth trips it.
Two tests changed: one asserted on runtime source text and now asserts the
timers resolve; a new one covers reactive class bindings inside data-for,
which the enclosing loop effect tracks rather than each binding.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The client runtime had no loop support, so any shared function using one
returned early -- Pagination and ButtonGroup were broken client-side, not
just in tests.
Adding loops exposed two further faults:
- A var reaching writeScope creates a signal and triggers a render sweep.
A declaration inside a function called during a render therefore looped
forever. Declarations now bind into the handler locals instead.
- A control block removed from the DOM keeps its effect in the renderers
list. Running it against a detached node threw, aborting the sweep and
leaving every later effect stale.
Also raises the reactive runtime budget to 53,000: the runtime had already
grown past 49,000 before this change, and 52,570 minified is 16,803 gzipped.
Two deferred minors: html-service leaves absent documentation undefined
rather than an empty string, and the extension declines tag auto-close on
multi-cursor edits rather than closing only the first cursor.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
{#if}, {#each} and their {:else}/{:else if}/{:empty} branches worked on
the server and after hydration, but a block nested inside another block
stayed empty once the outer block rerendered. Adding a row to a list
produced the row's markup with its inner block markers in place and
nothing between them, for the life of the page.
Two causes, both on the client-created path only:
reactive() registers an effect; effects run when renderAll sweeps the
list. A state change runs just the affected effects rather than sweeping,
so an effect registered during that rerender was queued and never
invoked. setupControlBlock now returns its runner and the creating block
invokes it immediately.
The first reactive pass is skipped so hydration does not discard
server-rendered DOM. A block created by a rerender has no server DOM, so
skipping its only pass left it permanently empty. firstRun is now keyed
off outerLocals, which is set only on the client-created path.
Verified in a browser as well as in tests: adding a group to a list now
renders the new row's nested {:else}, and the existing rows' nested loops
survive the rerender.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds a scratch page (kept intentionally, per controller ruling on Task 8)
and one end-to-end test that drives the real language server over LSP
stdio against that page's real text, verifying completion (with the
seo-block negative case tested via a simulated '<' keystroke and
mutation-verified against html-regions.ts), hover, folding ranges,
linked editing, and wrn/tagComplete all work together on realistic
content.
Regenerates routes.gen.ts and wrnexus.generated.d.ts for the new page's
route, required by check:generated-types.
Two checks from the original brief (auto-close-tag insertion and Emmet
Tab-expansion) require a live VS Code Extension Development Host and
are documented as outstanding manual verification in
.superpowers/sdd/2026-08-18-wrn-html-editing/task-8-report.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Replaces the direct-call-only test with an over-stdio test that exercises
server.ts's didClose handler itself, so it fails if the
clearHtmlRegionCache wiring is removed or misparameterized.
cli 0.8.42, csr 0.8.22, db 0.8.16, dev-server 0.8.38,
dev-toolbar 0.8.13, i18n 0.8.12.
Every previous version was already on the registry, so the HMR client
repair, the i18n JSON data block, the gateway WebSocket origin fix, and
the generated-dialect stamp were not reachable by consumers.
compiler and react are unchanged since their last publish and are not
bumped.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The WebSocket origin check compared the browser's Origin host, which
carries the port, against configured domains, which do not. publicOrigin
only ever matches domains[0], so every other domain fell through to that
comparison and was denied purely on the port: web.localhost:3000 never
matched web.localhost.
The result was a 403 on the HMR upgrade and a client reconnecting
forever, while the page itself loaded fine because HTTP routing resolves
the Host separately.
Compares hostnames now. Unrelated and lookalike-suffix origins are still
denied, and both cases are covered by tests.
Verified through a real gateway: the HMR socket opens on both localhost
and web.localhost, and a live edit reaches the browser.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The same generate command emitted ? one run and $1 the next, which looked
like non-determinism. It is not: postgres uses $1 placeholders where
sqlite and mysql use ?, and the driver comes from the active profile, so
building under a different profile rewrites this committed file.
The header now records the dialect it was generated for, making the flip
visible in the diff and explaining check:generated-types failures instead
of leaving them looking like random churn.
Worth deciding separately: a committed artifact whose contents depend on
the active profile will keep drifting. Either generate per dialect, or
stop committing it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
window.__wrnI18n was undefined in development: the payload shipped as an
executable inline script, and a document's CSP nonce is fixed at load, so
any such script arriving from a later response is blocked. Client
translations and language switching silently had no data.
The payload is now a type="application/json" block, which the browser
never executes and script-src therefore never applies to. The i18n
runtime, CSR navigation, and HMR all read the block instead of matching
window.__wrnI18n= with a regex.
Pages now render zero executable inline scripts, so an inline script-src
violation is structurally impossible rather than merely unobserved. Zero
framework JavaScript on island-free routes is unaffected: the block is
inert data, and nothing loads to read it unless the page needs it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A document's CSP nonce is fixed at load, so an inline script delivered by
a later response can never carry a nonce that document accepts. The HMR
client is now served at /__wrnexus/hmr-client.js, which script-src 'self'
already covers and which needs no nonce at all.
This removes one of the two inline scripts CSP was blocking in
development. The i18n data script is still blocked and needs the same
treatment; it is shared with the CSR navigation and HMR parsers, so
moving it spans @wrnexus/i18n, csr, and dev-server.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Pre-existing working-tree changes to migration SQL parsing, query
generation, and the packaging scripts. Committed as-is rather than
authored here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Excludes the toolbar's own bundle from the JavaScript budget, raises the
development thresholds, and skips WRNexus UI and theme stylesheets when
measuring CSS coverage, so unminified development modules and framework
styles stop reading as application problems.
Pre-existing working-tree change, committed as-is rather than authored
here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds gatewayWebSocketBackendHeaders so proxied upgrades carry application
identity while Bun keeps ownership of WebSocket framing.
Pre-existing working-tree change, committed as-is rather than authored
here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Component discovery is not utility-source discovery: scanning all
built-in and plugin component directories made Tailwind/Iconify generate
rules for components the app never renders. Packages that need scanning
opt in through styles.source.
Pre-existing working-tree change, committed as-is rather than authored
here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Raises the typescript devDependency across the workspace, bumps package
versions, re-adds ignoreDeprecations, and repoints the @wrnexus registry.
These were pre-existing working-tree changes, committed as-is rather than
authored here. The .npmrc change redirects @wrnexus publishes from
registry.npmjs.org to registry.workroot.in — confirm that is intended
before publishing from this branch.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The HMR client script was dead in the browser. HMR_CLIENT_JS is a
TypeScript template literal, so the regex [ \t\r\n] inside it was expanded
into real control characters, producing a regex literal containing a raw
newline — a syntax error that took the whole script down with "Invalid
regular expression: missing /". It now uses \s, and a test asserts the
emitted client parses and holds no control characters inside regex
literals; that test fails if the bug is reintroduced.
HMR also corrupted CSP nonces. A document's nonce is fixed at load, but
morph copied attributes from freshly fetched HTML, overwriting the live
nonce with one the browser will not honour. syncAttrs now leaves nonce
alone, and nodes moved across are re-stamped with the live nonce.
Islands vanished on every HMR update: morph puts the server placeholder
back over the mounted island. The island runtime now remounts on
wrnexus:hmr-updated. Remounting swaps the container for a bare clone —
re-rendering the existing root is a no-op once HMR has wiped the DOM
externally, and unmounting throws asynchronously because the nodes React
wants to remove are already gone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three bugs found by driving the dev server rather than reading code:
1. An island used inside a .wrn component still emitted a component mount
— only the page and nested-page render paths were covered.
2. Editing an island .tsx never rebuilt in dev. The bundle cache was keyed
on source path alone, and page modules are cached after the first
request so no compile runs to notice the change. The cache key now
includes mtime, and the file watcher rebuilds islands whose .tsx
changed.
3. A .wrn cache hit skipped island building entirely, so after a restart
with a warm cache no island bundle was ever produced. Island inputs are
now persisted beside the other artifacts and rebuilt on a cache hit.
The islands manifest is deliberately excluded from the artifact
completeness check: only the async compile path writes it, so requiring it
made the sync path miss the cache on every call.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The island pieces existed but nothing connected .wrn compilation to island
emission. Now:
- codegen emits a data-wrn-island placeholder for component tags bound to
.tsx imports, keeping .wrn components on the normal mount path
- the dev pipeline and static build resolve island imports, thread the
names into codegen, and build the bundles
- collectScripts adds /__wrnexus/islands.js only when island markup is
present, so island-free pages still ship nothing
- island routes classify as static-interactive via hasIslands
Three bugs found by driving a real page in the browser:
1. The mount runtime was never built anywhere, so the bootstrap 404'd and
no island mounted.
2. Building the runtime separately from the islands gave each its own copy
of React: "Cannot read properties of null (reading 'useState')". The
runtime is now an entrypoint of the same build so React stays in one
shared chunk. The existing single-React test only compared bundles
within one build and could not see across build boundaries.
3. Island props arrived as attribute strings, so start={3} was "3" and
incrementing produced "31" then "311". Props now follow JSX semantics:
{…} parses as JSON, quoted values stay strings, and a runtime
expression is a WRN-ISLAND-PROPS build error rather than a silent
wrong value.
island-codegen.ts no longer imports @wrnexus/core. Compiler modules are
bundled into the Node-only VS Code extension, which contains no other
packages, so a runtime import of core broke the editor compiler; the two
helpers are implemented locally and the core dependency is dropped again.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two guards protect the core promise: a route with no islands emits no
assets at all, and a page with several islands keeps React in one shared
chunk.
buildIslands now writes a generated entry per island instead of passing
component sources directly. Two islands sharing a source deduped to a
single entrypoint, and output order is not guaranteed to match input
order, so island names could bind to the wrong bundle.
Island modules are excluded from the editor compiler bundle: it globs
packages/compiler/src, and island-bundle.ts calls Bun.build while
island-codegen.ts imports @wrnexus/core — neither belongs in a Node-only
VS Code artifact.
Integration assertions share one build. bun test interferes with
Bun.build's module reads after several build calls in one process, while
the same calls succeed repeatedly outside the runner; production is
unaffected.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Disposes and re-creates island roots after a source change. Island state
resets by design; Fast Refresh needs a Babel/SWC transform plus a
runtime and is out of scope for v1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A route mounting an island ships JavaScript, so reporting it as zero-JS
static would make the framework's performance accounting wrong.
Adds a separate needsIslandRuntime flag rather than reusing
needsClientRuntime: an island needs the island runtime, not WRNexus's
reactive runtime, and conflating them would ship the wrong bundle.
analyzeRuntimeRequirements takes island presence as an optional second
argument, so existing callers are unaffected.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds /__wrnexus/islands.js (the bootstrap) and the /__wrnexus/island/
prefix (mount runtime, island bundles, shared chunks) to all three
serving paths.
Dev reuses the browserArtifactPaths registry pattern from pipeline.ts.
Prod mirrors the clientModulesDir handler, including its filename
allowlist, so island names cannot escape the output directory.
The bootstrap is inert without a data-wrn-island marker, so island-free
pages still download nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
splitting:true keeps React in one shared chunk so a page with several
islands does not ship react-dom repeatedly.
Island .tsx is compiled against React's JSX runtime via a Bun onLoad
plugin. The repo's root tsconfig sets jsxImportSource to @wrnexus/core,
so islands would otherwise compile to the HTML-string renderer and never
mount. A @jsxImportSource pragma only affects the file carrying it, so
injecting one into the generated entry is not enough — the injection has
to happen per source file. App-authored islands stay plain .tsx.
The JSX test asserts built output rather than generated entry text,
because the entry-text assertion passed while the mechanism did not work.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Mounts markers with client:only/load/visible/idle, and disposes roots on
route change so React roots, detached DOM, and store subscriptions do
not leak across client-side navigation.
Bundle load failures and malformed props JSON degrade to a warning and
leave the server markup intact rather than taking down the page.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ClintchizandClaude Opus 5