Conversation
stylex-build.ts runs @stylexjs/babel-plugin over every frontend file that imports @stylexjs/stylex ahead of the React Compiler pass and collects the compiled rules into one hashed sheet, linked after Tailwind's so a migrated style wins a source-order tie against the class it replaced. Both sheets run during the migration; Tailwind goes when the last utility does. tokens.stylex.ts bridges every color/radius/type/motion token to its base.css custom property, so html[data-theme] re-themes StyleX-declared styles with the stylesheets. typography.stylex.ts carries the eight-step type scale as shared styles. The 720px boundary stays authored once in lib/breakpoints.ts; StyleX keys spell the two media queries inline until cross-file const evaluation is trustworthy under Bun (guard test enforces the spelling).
Mechanical pass over every STATIC className="…" string: each class token becomes a stylex.create entry (names mirror the original tokens), variants land under pseudo/media keys, and the eight type roles move onto the shared scale in styles/typography.stylex.ts. Tokens that StyleX cannot express (data-[…] variants, group-*, arbitrary selectors, structural pseudos) stay as classes in markup for the residual stylesheet; component classes that were never utilities are kept verbatim. Dynamic className sites (cn(), template literals, ternaries) are untouched — they convert per-module next. The StyleX compile now chains inside the React Compiler onLoad pass so a converted file still gets compiler memoization; 521 files memoize and the bundle builds with both sheets linked during the migration. Codemod: scripts/stylex-codemod.ts (dry-run by default). Typecheck clean.
stylex-parity.test.ts pins the migration invariants: one 720px spelling everywhere, tokens resolving through base.css, a raw-color ratchet inside stylex.create blocks, and (as a failing-until-waves-merge gate) every static className token resolving in a shipped stylesheet. styles/residual.css carries the compiled Tailwind rules StyleX cannot express, verbatim from the sheet, for the classes that remain in markup; scripts/stylex-residual.ts regenerates it. The @stylexjs/stylex patch makes an uncompiled create() return its input instead of throwing, so bun test can import converted components: component tests assert behavior, while class fidelity stays with the build and the parity guards. Without it every converted import throws at test time.
# Conflicts: # packages/core/opensession-server/src/frontend/components/GithubManifestSetup.tsx # packages/core/opensession-server/src/frontend/ui/tooltip.tsx
# Conflicts: # packages/core/opensession-server/src/frontend/components/Feed.tsx # packages/core/opensession-server/src/frontend/components/Prs.tsx # packages/core/opensession-server/src/frontend/components/UserPicker.tsx # packages/core/opensession-server/src/frontend/components/settings/IngressPanel.tsx # packages/core/opensession-server/src/frontend/components/sidebar/HoverCards.tsx
# Conflicts: # packages/core/opensession-server/src/frontend/components/Feed.tsx # packages/core/opensession-server/src/frontend/components/GithubManifestSetup.tsx # packages/core/opensession-server/src/frontend/components/Prs.tsx # packages/core/opensession-server/src/frontend/components/settings/IngressPanel.tsx # packages/core/opensession-server/src/frontend/components/sidebar/SetupWidget.tsx
# Conflicts: # packages/core/opensession-server/src/frontend/components/sidebar/SetupWidget.tsx
# Conflicts: # packages/core/opensession-server/src/frontend/components/Feed.tsx # packages/core/opensession-server/src/frontend/components/GithubManifestSetup.tsx
# Conflicts: # packages/core/opensession-server/src/frontend/components/sidebar/SetupWidget.tsx
# Conflicts: # packages/core/opensession-server/src/frontend/components/Feed.tsx # packages/core/opensession-server/src/frontend/components/FirstMile.tsx
main moved 457 commits ahead while the port was in review, with heavy frontend refactors (SessionViewer, Sidebar, App shell, PR review pages). This merge takes main's newest logic everywhere and reruns the full conversion machinery over it so the port keeps parity with main's UI: - Rebuild utility-compat.stylex.ts from main's current compiled sheet (1805 entries) and regenerate residual.css with the specificity bridge; container queries and --tw plumbing utilities stay residual. - Re-apply the StyleX build wiring (frontend-build.ts) on top of main's stable-shell/preload machinery; sxName replaces twName in BundleMeta. - Restore main's base.css deltas lost in resolution (code-well split, reasoning shimmer contrast, review-code themes, settings drag rules, composer image attachment rule) so token parity with main is exact. - Port main's real deltas the mechanical resolution dropped: ReviewToolbar transparent outer surface with smooth-shadow-ring-sm, WSClientMessage typing on SelectionToSession/GitStatus send props. - Make cn()/utilityClassName keep source spellings visible under uncompiled StyleX (bun test) so suite tests read the tokens main was written against; compiled builds keep hashed classes. - Extend normalizeQuery for Tailwind v4 range syntax (lg: variants), teach stylex-compat-rewrite to wrap plain-string JSX attributes with expression braces, and fix the residual specificity guard to judge the whole rule (formatter wraps selectors/bridges across lines). - Adapt main's new test assertions to StyleX semantics (inline styles, compat spellings, registered override markers) without weakening their intent; remaining suite failures match main's own pre-existing isolation quirks exactly, with the port adding 13 passing tests. check: format, typecheck, lint, isolated unit tests, snapshots, module side effects, and a production build with only the StyleX sheet (no Tailwind asset) all pass.
The three incremental main merges brought heavy Tailwind UI work (expandable workspace-summary changes with per-file diff previews, code-workspace checkout preferences, phone setting-row constraints, review chrome) along with formatter churn that exposed three pipeline gaps: - stylex-residual.ts parsed the compat map one entry per line; oxfmt wraps long key/value pairs, so make the parser tolerate whitespace between the key colon and its sx binding. - Rebuild utility-compat.stylex.ts against main's current compiled sheet and add every sheet token to the request set (absolute/border/contents were never requested and surfaced as raw classes in follow-up merges). - Re-apply the StyleX adaptations for WorkspacePane.test.ts source assertions (norm() whitespace helper, whitespace-flexible regexes for the compactToolbar ternaries, reviewStateClass/LoadingState, review toolbar sx tokens, and the TopBar indexOf slice) that main's newer test version replaced. check: format, typecheck, lint, isolated unit tests, snapshots, module side effects, and a production build with only the StyleX sheet all pass; the full frontend suite failure set is identical to main's own.
main's latest review changes moved the per-file header backgrounds onto --review-file-header-* custom properties with Vercel-style tone mixes and a new non-sticky overflow surface. Re-run the conversion pipeline over the merged Tailwind sources and adapt WorkspacePane.test.ts source assertions to the StyleX forms (utilityClassName ternaries with optional trailing commas, sx toolbar tokens, the norm() whitespace helper). Also regenerate the compat map with a fresh sheet-token request set: the earlier refusal loop had dropped absolute/border/contents even though the codemod converts them, leaving raw classes in ROW_ACTION. check: format, typecheck, lint, isolated unit tests, snapshots, module side effects, and a StyleX-only production build all pass; the frontend suite failure set remains identical to main's own.
oxfmt rewrites "absolute": sx.absolute to absolute: sx.absolute when the key is a valid identifier. The residual script's key parser only matched quoted keys, so those entries silently vanished from the compatibility set and the gate reported convertibles that the runtime map actually resolves. Accept both spellings (and fix the probe-vs-batch confusion in the request-set rebuild while at it).
Contributor
Author
|
Brought the port up to date with main (was 457 commits behind, main kept moving during the update; final state merges through What landed
Pipeline bugs found and fixed
Verification
Started by Jaap Frolich in this OS session |
Contributor
🤖 OS review · request changes · confidence 2/5Safe once the two P1 regressions below are fixed. The StyleX migration is broad and includes fail-closed build metadata and parity guards, but it breaks the gateway’s stable-shell handoff and leaves the standard OPENSESSION_DEV preview serving stale frontend assets after edits.
🔁 Not merge-ready and no live session owns this branch — add the |
![open-session-os-tella-dev[bot]](https://avatars.githubusercontent.com/u/65127121?s=60&v=4){kind=small}
| publishStableShell(true); | ||
| return Promise.resolve(); | ||
| } | ||
| if (frontend.version) return Promise.resolve(); |
Contributor
There was a problem hiding this comment.
🔴 P1 — Republish the preloaded frontend after gateway activation
A standby calls preloadPreparedFrontend() before its activation fence, where applyBundle() deliberately cannot publish because OPENSESSION_GATEWAY_ROLE is still standby. After activation, opensession.ts changes the role to active and calls ensureFrontendBuilt(), but this early return now skips publishStableShell(true). Consequently stable-frontend.json remains pinned to the previous release. During the handoff or a later backend gap, the supervisor serves that old shell; a browser opened then can retain an old client against the newly activated backend, including incompatible API contracts. The previous implementation explicitly republished here, and this PR also removes the regression test covering that sequence. Restore the version-present branch that calls publishStableShell(true) before returning.
| // ours to keep fresh, so drop it and let the next request recompile. | ||
| devTailwind = null; | ||
| // Dev serves the source tree through Bun's HMR server; StyleX compiles | ||
| // per request there, so there is no cached sheet to drop. | ||
| return; | ||
| } |
Contributor
There was a problem hiding this comment.
🔴 P1 — Keep OPENSESSION_DEV previews rebuilding after frontend edits
The standard preview path (.agents/start.sh) starts the server with OPENSESSION_DEV=1. This PR replaces Bun’s source/HMR entry with a one-time compiled bundle, while opensession.ts still starts its frontend watcher only under !IS_DEV, and this branch makes rebuild requests no-ops. For example, after opening Preview and editing Composer.tsx, every reload continues serving the bundle compiled at process startup until the entire preview is restarted. That breaks the documented self-development loop and contradicts the comment that StyleX compiles per request. Add an OPENSESSION_DEV watcher/rebuild path for the compiled StyleX bundle, or retain an actual StyleX-capable HMR path.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Verification
bun run typecheckbun run lint(0 errors; 163 existing warnings)bun test packages/core/opensession-server/src/frontend(1400 pass, 0 fail, 6019 assertions)bun scripts/check-module-side-effects.ts(416 modules clean)compileAssets()(486 files;styleEngine: stylex-v1; no Tailwind asset)Started by Jaap Frolich in this OS session