feat(extensions): match file languages by filename or glob

[benvinegar]

Summary

• extend hunk.registerFileLanguage with declarative exact-filename and basename/path glob matchers while preserving the existing extension-string shorthand
• resolve selectors in Hunk with documented precedence, platform-neutral slash semantics, and protected .mts/.cts mappings
• pin the selected language into Pierre metadata so non-extension selectors render consistently
• replace selector generations atomically across extension reloads and restore the previous generation when a reload fails before commit
• publish the additive contract as extension API v9 and document the new matcher forms

This provides the extension seam needed to move mappings such as Bazel/Starlark filenames out of core in a follow-up to #813 without exposing arbitrary regular expressions.

Example use cases for extension implementers

export default function activate(hunk: HunkExtensionApi) {
  // Existing shorthand: map a dotted extension, case-insensitively.
  hunk.registerFileLanguage(".bzl", "python");

  // Bare tool/config filenames at any directory depth.
  hunk.registerFileLanguage({ kind: "filename", value: "BUILD" }, "python");
  hunk.registerFileLanguage({ kind: "filename", value: "WORKSPACE" }, "python");

  // Filename families, regardless of their containing directory.
  hunk.registerFileLanguage(
    { kind: "glob", value: "Dockerfile.*", target: "basename" },
    "dockerfile",
  );

  // Generated or vendored trees, matched against the complete review path.
  hunk.registerFileLanguage(
    { kind: "glob", value: "generated/**/*.proto", target: "path" },
    "protobuf",
  );
}

This lets extensions support conventions that cannot be represented by a final extension alone: Bazel/Starlark files such as BUILD, Dockerfile variants such as Dockerfile.dev, framework-specific filename families, and files whose language depends on where they live in a repository.

String registrations remain extension-only shorthand: use { kind: "filename" } for a literal bare filename. Filename and glob selectors are case-sensitive. target: "basename" sees only the last path component, while target: "path" sees the decoded review path using / as its only separator; backslashes remain literal filename characters.

Validation

• bun run typecheck
• bun run lint
• bun run deps:check
• bun run check:docs
• bun run check:pack
• bun run test — 3,054 passed, 10 skipped
• bun run test:integration — 128 passed
• bun run test:tty-smoke — 9 passed
• real TTY smoke run against a TypeScript diff
• all GitHub checks, including Windows compatibility

This PR description was generated by Pi using gpt-5.6-sol

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
hunk-web	Ready	Preview	Aug 26, 2026 3:41am

[greptile-apps]

Greptile Summary

The PR expands extension file-language registration from extension suffixes to exact filenames and basename/path globs, while preserving deterministic precedence and protected TypeScript mappings.

• Adds and publishes extension API v9 matcher types, validation, and documentation.
• Introduces versioned, atomically replaceable selector generations with rollback during failed reloads.
• Resolves selectors while constructing diff files and pins the selected language into Pierre metadata.
• Adds unit and integration coverage for matching, precedence, reload replacement, rollback, and public API robustness.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete blocking or independently actionable non-blocking issues identified.

Matcher validation, precedence, metadata propagation, generation replacement, and reload rollback are implemented consistently and covered across unit and host-level tests.

Important Files Changed

Filename	Overview
src/core/changeset/fileLanguageLookup.ts	Compiles versioned selectors lazily and implements reserved-extension, filename, glob, and longest-extension precedence.
src/core/changeset/fileLanguage.ts	Replaces the prior pending-map model with immutable selector snapshots and atomic generation replacement/restoration.
src/extensions/runExtension.ts	Validates and canonicalizes the new public matcher forms while preserving extension-factory failure isolation.
src/extensions/apply.ts	Applies complete selector generations and continues to reject direct attempts to override reserved extensions.
src/app/sessionBootstrap.ts	Captures and restores the prior selector generation when candidate session bootstrap fails.
src/ui/AppHost.tsx	Integrates selector rollback into live reload commit, shutdown, and publication-failure paths.
src/core/changeset/diffFile.ts	Resolves each file’s selected language and pins it into renderer metadata for non-extension matchers.
src/extension-api/types.ts	Publishes the additive matcher union and advances the extension API generation to version 9.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Extension factory] --> B[Validate matcher]
  B --> C[Extension registry]
  C --> D[Replace active selector generation]
  D --> E[Load changeset]
  E --> F[Resolve filename, glob, or extension]
  F --> G[Pin language into Pierre metadata]
  E -->|Failure| H[Restore previous generation]
  G --> I[Commit review and extension generation]

Loading

_{Reviews (1): Last reviewed commit: "feat(extensions): match file languages b..." | Re-trigger Greptile}