solid

Summary Follow-up to #3041 — the companion backfill itself is fixed in rc.2 (verified against the published tarball), but the reported "still one value behind sometimes" turned out to be a distinct, long-standing bug (reproduces on rc.1, rc.2, and ). Symptom: content whose async read is short-circuited behind a pending gate — — lands showing the previous committed value and never corrects until the next write. A flat effect that reads unconditionally is fine, and user-tier is unaffected, which is why it only bites "sometimes". Root cause When the pending gate flips off during the landing flush, the render effect re-runs and reads the async memo for the first time. That read happens under the isPending companion's optimistic lane ( assigns the source's , and settle keeps it), so serves the committed old value. The staged then promotes via , which deliberately never re-notifies — stranding the reader on the old value. The engine already has a replay contract for exactly this shape (gated subs recorded at read time, replayed by after commit — #2963), but the branch of only recorded when no** transaction was active, so transaction landings fell through the hole. Fix In , whenever a committed-view read hides a staged , record the reader into . The existing replay re-runs it right after promotion, within the same synchronous flush — the effect sequence goes with no paint in between. Tests — signals-level regression pinning the exact value sequence (fails before the fix with the effect stuck on ). — the DOM shape through `pikachucharizardpackages/solid-web/test/latest-ispending-issue-3041.spec.tsx` — control spec pinning the full #3041 app shape (banner + content across sequential transitions and rapid mid-flight clicks), which passed before and after. Full suites green: solid-signals 1351, solid-web 425, solid 561. 🤖 Generated with Claude Code

Summary Lift the JSX compilers as and , and collapse the DOM/SSR runtime into (plus / / ) so Solid 2.0 no longer depends on a sibling checkout. Fold render/hydrate policy into the runtime, migrate the absorbed DOM code to TypeScript feature folders, and bake Solid defaults (, ) into both compilers. Make JS CI pass on without a native compiler build: stop tracking the generated Babel bundle, skip in root , and fix HydrationScript / published JSX identity so examples typecheck. File compiler/runtime issues on this repo after merge, not ryansolid/dom-expressions. Binary publish stays a separate dispatch workflow, same as expressions. Test plan [ ] JS CI green on this PR ( + ; compiler native matrix is push-to- / dispatch only) [ ] Confirm no remaining package.json dependencies [ ] After merge, add a redirect/banner on the expressions repo so new issues land here Made with Cursor

Summary move the server-function implementation and declarations into expose Solid-owned protocol metadata and flash helpers from the core entry build the client, server, and rich-argument entries from Solid-owned source inject the server-function protocol into the generic dom-expressions frame adapters move the complete protocol test suite into Solid preserve the existing public API and generated package output The runtime source is synchronized with dom-expressions at . Why is the framework-neutral rendering layer. Server functions are not rendering primitives. They define Solid's client/server protocol, including reference registration, request encoding, dispatch, errors, redirects, revalidation, CSRF, progressive enhancement, and observability. The compiler and public API already identify this as Solid functionality through . Keeping the implementation in dom-expressions makes Solid's protocol an implicit responsibility of a lower-level renderer and exposes Solid-specific concepts to every dom-expressions consumer. This change aligns implementation ownership with the existing public boundary. Generic framing and serialization primitives remain in dom-expressions, while Solid supplies the server-function protocol adapter. The dom-expressions copy remains temporarily so this can land without an atomic cross-repository release. Its removal is prepared in ryansolid/dom-expressions#577. This PR has been tested against both the current dom runtime and #577, so no third migration PR is required. Verification full workspace build 1,042 tests passed, 2 skipped build, declarations, and type tests against the current dom runtime the same build, declarations, type tests, and runtime tests against dom-expressions #577 ESM import and packaged-type integration tests npm package dry run CodSpeed CodSpeed could not find a successful run for the actual base at , so it compared this branch to older across different runtime environments. The reported changes are all signals and store benchmarks, which this PR does not modify.