Feat/svelte extraction binding - #84
Merged
Merged
Conversation
…xtraction-only system binding The Svelte arc, rebased clean onto current main (replaces the local merge-based chain, preserved at backup/main-with-bench): Usage-extraction poc — shared raw-source ingestion boundary (source-ingestion.ts) with the native Svelte source adapter projecting .svelte usage facts into the OXC pipeline; framework-neutral class-resolver runtime entry (@animus-ui/system/class-resolver) with runtimeImport plugin override; Svelte 5 e2e canary (client + SSR builds, generated-CSS and runtime-HTML assertions); dual outside-repository packed consumers (resolver-only production + full authoring). React-absent authoring (svelte-extraction-only-system-binding, inc 01): - D1: resolver definitions bind the configured system through a type-only import + declared chain root under required strict extraction; the application module graph never evaluates the React-reaching builder barrel. - D2: the Vite geological reset snapshots replacement plans, diffs them after successful publication, and evicts changed source modules across every environment graph before the full reload; focused falsifier tests cover changed/equal/failed-reset arms with a demonstrated red. - D3: React removed from the Svelte authoring manifests (exact @types/react retained per DEF-2); packed verification proves React undeclared/absent/unresolvable through strict TS, a specifier- rewrite-aware middleware-mode dev transform probe, client/SSR builds, and runtime assertions. Also: parity corpus gains the two Svelte integration units (refreshed under svelte-usage-extraction-poc-corpus-20260809 against the transform-sources oracle); packages/_bench/ and .vite-hooks/ are now gitignored local tooling. Gates: verify:packed green end-to-end, G2 probe exact, G3 focused 16/16, compile / 1282 TS units / 165+1 integration / 16 dependent-owner tasks green; subagent cold review findings dispositioned. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…gestion arc Incremental quarantine parity (F4/F5/F6): withoutInvalidOriginals moves to the shared pipeline and every incremental path — hmr edit, delete prune, transform detection, rediscovery fold, vite geological reset, next incremental batch — quarantines invalid originals per-file exactly like buildStart instead of aborting the whole re-analysis on any corpus diagnostic (which froze HMR on default config: .mdx in DEFAULT_EXTENSIONS with the optional peer absent). Deletes never restore pruned entries (one watcher event means a restored entry is a permanent ghost), the .mdx fold guard is back, warns dedupe per (path, line) and clear on clean republication, and next's misleading 'idle' status write on abort is gone. Advisory parse diagnostics (F1): recovered OXC parse notes warn in every mode, never strict-throw, never quarantine — extraction must not be stricter than the host bundler (JSX in .js builds again; full JSX-in-.js support via analysis-path rewrite deferred: it moves plan keys). Resolver index (F2 rejected, F3 fixed): renamed re-exports KEEP failing closed — live probes against the real engine show usage attribution follows same-name barrel hops but not renames, so witnessing renames would silently lose usage; the boundary now carries its engine-layer proof test and true rationale. NodeNext relative specifiers with emitted extensions (./definition.js) resolve to their source modules, proven end-to-end through engine pruning. Adapter fail-open shapes (F7/F9/F10): the callable string form witnesses identically to .attrs() under the same argument/access rules; module- script imports are in scope for instance-script calls with the import copied into the projection; the template fragment is scanned (dedicated walker — Fragment containers carry no spans) and resolver calls there fail closed with SVELTE_ATTRS_TEMPLATE_UNSUPPORTED. Host coherence (F11/F12/F13/F14): dev fileCache seeds from the accepted corpus before the analysis gate (a failed non-strict buildStart no longer strands HMR with a one-file universe); runExclusiveAnalysis serializes transform detection, hot updates, and geological resets per context; the external-token join runs in publishSourceIngestion after generated-child owners land (next-plugin ordering parity); a runtimeImport override throws in every mode when any replacement needs createComponent/ createComposedFamily, naming the offending components. Parity honesty (F15): the integration corpus enumerates .ts and refuses zero-file units — svelte-usage gains its real chain surfaces, the hollow svelte-lifecycle unit leaves the corpus (65/65), refreshed under svelte-parity-corpus-enumeration-20260810 with the register cycled back to empty. F8 documented as the inline-literal pruning rule; dead buildFileEntriesFromCache removed. Gates: compile all packages, 1293 TS units, integration (incl. live barrel-boundary and NodeNext probes), G3 focused 18/18, canary, parity 65/65 both modes, 16 dependent-owner tasks, packed end-to-end, lint/fmt. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…es JSX-enabled
Renamed re-export attribution (lifts the svelte fail-closed boundary):
resolve_usage_identity walks renamed chains to the defining binding —
sourced re-export hops via follow_reexports, defining-module local
renames (export { badge as fancyBadge }), and import-then-local-export
barrels (import { badge } …; export { badge as pill }), cycle-guarded,
alternating until a chain id lands. A renamed consumer's usage now
reaches the chain, so another consumer's literal can no longer prune the
variant a renamed import renders. The integration falsifiers caught the
import-then-local-export shape missing from the first implementation
before it could ship as silent fail-open. TS-side, the resolver index's
name-equality gate is lifted (renames classify as resolvers), the
fail-closed tests flip to witnessing, the live boundary test proves
renamed pruning end-to-end through the real engine, and the guide's
boundary section moves renames into the supported set.
JSX in .js: source_type_for parses .js JSX-enabled — the Babel/SWC/
esbuild ecosystem norm (CRA / legacy-React / Next pages/*.js). This is
the deliberate alternative to the analysis-path rewrite, which would
have churned replacement-plan keys for every .js consumer; recorded as
an explicit deviation from the standing TS-side-rewrite preference,
whose premise (rewrite is the cheap path) inverts here. JSX grammar
activates only at expression start, so plain-JS comparison chains are
unaffected (covered). Recovered parse diagnostics remain advisory for
genuinely malformed sources.
Gates: extract-v2 553 units (renamed hops red-checked via removal),
clippy -D warnings, cargo-machete, canary, parity 65/65 both modes with
no baseline drift, integration 166+1 todo incl. flipped renamed-barrel
receipts, TS units 1293, compile, lint/fmt.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…d session spine The CLI commit moved the session spine from next-plugin/src to packages/extract/session and replaced per-test vi.mock singleton stubs with the setEngineApiOverride seam; the gauntlet commit replaced raw exclude patterns with ExcludeMatcher. This merge carries pr-84's source ingestion (ingestRawSources / surfaceSourceDiagnostics / publish) onto the moved spine, keeps the seam-based test injection with an added extractFacts stub, and adopts ExcludeMatcher at both resolveScanConfig call sites. shouldHandleMdx/missingDepFlag locals dropped: MDX preprocessing now lives inside ingestSourceEntries. bun.lock regenerated from the merged manifests. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…errides The merged session requires extractFacts for source adaptation; these three files' seam overrides predate pr-84 and auto-merged without it. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The vi.mock targeted the pre-hoist '../src/singleton' path deleted by the session move; the seam override preserves the test's conditional extractFacts-absent arm, and engineApi now resolves through the moved singleton (the seam key survives resetAnimusGlobals by design). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…on per host Hoists the forked surfaceSourceDiagnostics / ingestRawSources policy into createSourceIngestor (pipeline-owned: capability guard, warn-dedupe lifecycle, per-host prefix; restores the prefixed-warn parity next-plugin had drifted from) and adds a host-owned facts memo so incremental corpus passes re-extract only changed entries instead of re-parsing the whole project through the native boundary on every HMR/watch event. Collapses the ingest -> quarantine -> analyze -> publish transaction, previously open-coded at 8 call sites with five publish-guard spellings, into PluginContext.analyzeIngested (beforeAnalysis hook carries the two documented ordering constraints) and the session's ingestAccepted core. The context probe mirrors the real transaction over its overridable parts. hmr's per-event O(corpus) entry scan is now an O(1) analysisEntryCache read. addAnalysisEntry/addAdaptedEntries became closures; identity entries reuse the original's hash instead of re-hashing the corpus every pass; ResolverExportIndex memoizes attribution and relative resolution. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
svelte-source-adapter: byte offsets take an ASCII fast path (was a prefix copy per span endpoint — quadratic per adaptation); import bindings compute once per program instead of twice each for module and instance; script/fragment AST walks skip entirely when no candidate import binding exists; svelte/compiler loads through a success-memoized dynamic import (missing-peer still retries, so mid-session installs recover). createClassResolver: the callable string form resolves classes directly instead of building and discarding the attributes object and its serialized style string on every render. discover-packages: preprocessFile survived the move of MDX adaptation into ingestSourceEntries with both callers passing identity — replaced by the optional onSourceRead observer, which names what the session actually does (record raw-file identity); the null-skip and path re-key branches retire with it. extraction-session: the 11-field snapshot/restore parallel lists become snapshotSourceState/restoreSourceState — one field list, Object.assign restore that cannot omit a member. Removed: PluginContext.analysisEntries() (zero callers), SVELTE_RUNTIME_SPIKE.md (superseded tombstone pointing at a gitignored path; README now points at the committed e2e fixture). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The guide previously lived only at docs/svelte.md, which the bare 'docs' gitignore entry swallows — unreachable in-tree and linked from README as a dead path. It now ships at /docs/advanced/svelte in the showcase (nav entry + MDX page; the content glob picks it up), and the README points at the guide plus the e2e/svelte-app working consumer. Verified: page renders at /docs/advanced/svelte with zero console errors (all four sections, six highlighted fences, strict-mode callout, sidebar entry) and the showcase owner claim (build + assert + Wrangler dry-run) passes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
1 participant
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.
No description provided.