Skip to content

feat(C/N2): exact source attribution for loaded PGF/TikZ imports - #55

Merged
i-shutov merged 4 commits into
mainfrom
feat/c-n2-import-evidence-sites-20261031
Oct 31, 2026
Merged

i-shutov merged 4 commits into
mainfrom
feat/c-n2-import-evidence-sites-20261031

Conversation

@i-shutov

@i-shutov i-shutov commented Oct 31, 2026 •

Copy link
Copy Markdown
Owner

Scope and baseline

Independent stream C only. Exact base main 6d15d86344ab1281391d3ea1b458dc18ae443985 (N1 #44 and N2 #49/#51 already merged). Current exact head 0cc7427cbbdee8b5ffe5fa63dcc8737ae5781cba. No old #39 transplant. No A/B/D/E/F edits.

User-facing vertical

  • ProjectImports::declaration_sites(ImportKind, name) -> Vec<DeclarationSite>, with exact UTF-8 name and physical-line spans for trusted direct \\usepackage / \\usetikzlibrary declarations. B's inspector can focus/highlight the real evidence instead of guessing dependencies; there is no new edit authority.
  • Preserves package options, comma lists, indentation, BOM, CRLF and Cyrillic comments. Conditional/opaque/macro-dependent imports have no fabricated source provenance. Positive direct literal evidence before later unrelated macros remains available.
  • Corrects a real false-positive: \\usepackage{tikz} before \\documentclass could previously be called CONFIRMED after a later class. Such invalid preamble order is now UNKNOWN and N2 source patches refuse.
  • Existing N2 inspect/status/plan_import/plan_library_disconnect, SourceTarget/SourcePatch and six N1 generators, Russian help, five independent state dimensions are preserved.

QA and evidence

Seven new Rust integration tests for exact offsets, source attribution, negative/unknown cases, duplicate imports, invalid order, and all-six regression. Code, tests, C-owned contract, plans/insights updated together. Local cargo/rustc unavailable: NOT_RUN. Initial SHA 2d81c9 FAILED exact hosted rustfmt (native CI 41205816101); 9 original formatter hunks were applied in b4e23ab, and the subsequent host Clippy too_many_arguments warning was corrected in 0cc7427c by reducing the private recorder to 7 arguments, without #[allow]. Current exact-head CI is queued/in progress; NO PASS CLAIM until it finishes. A TeX compilation or hosted Windows Server check is not physical Win10/MiKTeX/M1/M4 GUI acceptance.

Handoff

B PR #50 notified in advance (additive read-only API, no type rewrite); B owns source-focus UI and visible diff/Undo. D/E/F independent tests/help/source QA remain in their respective scopes.

Exclusions

No TeX parser/CST, macro expansion, geometry solver, PDF decompilation, silent package installation, engine switching or GUI implementation.

Exact-head QA checkpoint — 2026-10-31 (0cc7427cbbdee8b5ffe5fa63dcc8737ae5781cba)

  • Native CI SUCCESS 3/3: Linux XeLaTeX fixture, hosted macOS 15 and Windows Server 2022 Rust jobs.
  • Independent D QA SUCCESS 7/7: hosted Ubuntu/macOS/Windows Server Rust fmt/tests/Clippy, actual XeLaTeX and independent artifact gate; includes seven new C-owned attribution tests.
  • N1 actual app state → TikZ source → external XeLaTeX SUCCESS; font/glyph integrity SUCCESS.
  • Distinct MiKTeX 25.12 hosted Windows Server job was IN_PROGRESS at this checkpoint, not PASS. Windows Server CI is never physical Windows 10/MiKTeX acceptance.
  • Only five C-owned files changed; no A/B/D/E/F edits. Physical Win10/MiKTeX, M1/M4 GUI and B inspector source-focus are NOT_RUN. This is read-only source declaration provenance, not GUI feature/release acceptance.

Ready for independent code review at the exact verified HEAD; integrator/B own adoption and merge, no automatic merge requested.

Copy link
Copy Markdown
Owner Author

C/N2 exact-head QA checkpoint, 2026-10-31, commit 0cc7427cbbdee8b5ffe5fa63dcc8737ae5781cba:

Copy link
Copy Markdown
Owner Author

Post-review baseline drift note (2026-10-31): while reviewing PR #55, main advanced beyond initial base 6d15d86344ab1281391d3ea1b458dc18ae443985 through integrator documentation PR #54 (observed 994b4751999fa0a3711a27b80c844ca38307a295 and later commits). compare(main...feat/c-n2-import-evidence-sites-20261031) still lists only the same five C-owned files in the PR diff; no A/B/D/E/F production files or the integrator's documentation are overwritten. Exact-head 0cc7427 CI is green as documented, but CI for a future integrated merge SHA is NOT_RUN; integrator should refresh the base/merge queue, re-evaluate mergeability and run its post-merge gates rather than transposing old #39. No force push or cross-lane edits performed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant