[i-shutov]

Large integration block: #10 conflict resolution + #21 CI activation

Starting main e3a78ae5646ef24994161c8ebe7bfaaf7a7d2ab6 (after #23, #20, #21 merged). Reuses proven A #10 content without forcing its conflicting diagnostics branch.

Code

• XeLaTeX BuildOptions optional durable .log, cancellation/timeout log retention, source-filename safety, dotted output names, normalized .fls and source/line diagnostics.
• CLI build source.tex output.pdf [--log full.log] with atomic PDF publish.
• Reconciles F02 font warnings (Missing character, LaTeX Font Warning and fontspec warnings) with A's file/line diagnostics.
• egui shows parsed failure log/line, permits line jumps only when the logged file matches the open document, and retains stale preview on failure.
• Unix fake-compiler tests and CLI source-preservation tests.
• Linux real XeLaTeX and three-host D QA invoke C03: run all five installed-package fixture records, independently verify actual source/log/PDF hashes, 1-page PDF status and expected Lua-only engine negative, persist audit JSON.
• Updated plans/, insights/, architecture/extension docs and CHANGELOG.

Evidence and gates

A upstream SHA 3af867cf was green on 3-job CI, but not directly mergeable after F02. C03 #21 upstream green on 3 workflows, but standalone auditor was not invoked in CI. This integration must prove fresh exact SHA via Rust fmt/check/test/clippy, 3 native CI builds, six D QA jobs, independent font proof and uploaded audit artifacts. Until then DRAFT / NOT VERIFIED.

No physically verified Windows 10 22H2 or user M1/M4 GUI tests. No claim of complete visual TikZ editing or Lua graphdrawing under XeLaTeX.

[i-shutov]

D04 independent QA review (exact source head 1db951c02d7aa7a896a661d7f2c17bd7de91c973, 2026-10-31): blocking Windows regression.

Verified Actions run 41180076260: 5/6 D jobs PASS, Windows Server 2022 CLI→XeLaTeX FAIL (job 124475503503). Downloaded uploaded Windows artifact 13035155032, inspected its actual qa-report.json and stdout/stderr: 5 positive fixtures FAIL with ! Undefined control sequence. <*> .\\quickstart / .tex (equivalently .\\main / .tex), exit=3, no PDF. Source hashes unchanged. The new runner is passing a Windows .\\foo.tex operand back to XeTeX, which treats \\foo as a TeX control sequence; this is the previously identified #10 regression, not a CTAN/font installation error. Do not treat GitHub main CI 3/3 success or F01 1/1 success as overall PASS.

Additional D gate hole: current g08 intentional TeX-negative fixture reports PASS on Windows despite exactly the same invalid operand! Its log lacks the fixture's actual \\TikzariumIntentionallyUndefinedCommand. Existing D test only requires exit=3, XeLaTeX exited, no PDF; so any failing TeX process can masquerade as expected. I am preparing a D-only test harness change to require the specific undefined-command diagnostic, not editing your A-owned runner.

Downloaded and independently inspected Linux artifact 13034562992 and hosted Mac ARM64 13034647150. For each, all five extension observation.toml parse; source digests match repo source, console+TeX log+4 PDFs hashes match, all 4 positive PDFs are one-page and without Missing character:, Lua-only negative has nonzero exit + full You need to run LuaTeX to use the graph drawing library. (internally wrapped grap\nh on Linux, gra\nph on Mac) and no PDF. Both extensions/audit.json: accepted=true,complete_suite=true,verified_fixtures=5,errors=[]. These are CI_ONLY, not physical Win10/M1/M4 acceptance.

A/I05 bounded fix requested: restore portable literal ./ prefix instead of platform Path::join(".", filename) for XeTeX argv; add a Windows-targeted regression and rerun exact-head 3+6+F01; do not merge while D Windows remains red.

[i-shutov]

I05 integrator fix / exact-head gate (2026-10-31)

I confirmed base main e3a78ae5646ef24994161c8ebe7bfaaf7a7d2ab6 and old I05 head 1db951c02d7aa7a896a661d7f2c17bd7de91c973. Its D QA 41180076260 was FAIL 5/6: Windows Server 2022 returned TeXFailure 3 for five expected-valid fixtures. This is not Win10 evidence and G08's incidental PASS is not a valid source-specific negative proof (see #27).

Pushed new head 230428d6e7de0c4c75ca73b9fc917cbda282346e by an expected-head ref lease. Fix in build.rs: preserve original OsStr filename and give XeLaTeX a literal ./ prefix, not Windows-native .\\ path which TeX treats as a control sequence. Added Unicode/dash-name unit regression and updated CHANGELOG / I05 plan / QA insights. No TeX source was edited.

Exact-head checks started: Rust CI 41182303483, D QA 41182303487, font proof 41182303497. Do not merge while queued/running. Require 3/3 + 6/6 + 1/1 PASS and inspect actual C03 observation/audit JSON on all three hosted systems; then acquire fresh main/head lease. After successful integration, close #10 as superseded, not force-merged. Physical Windows 10 x64 / Mac M1/M4 GUI acceptance remains NOT_RUN.

[i-shutov]

I05 exact-head ACCEPTED for hosted integration (2026-10-31)

Head: 230428d6e7de0c4c75ca73b9fc917cbda282346e on feat/i05-runner-diagnostics-extension-audit with base e3a78ae5646ef24994161c8ebe7bfaaf7a7d2ab6. All required GitHub-hosted checks completed SUCCESS on this exact head: Rust CI 41182303483 3/3, D QA 41182303487 6/6, font proof 41182303497 1/1. This includes actual Rust CLI→XeLaTeX+Poppler on hosted Windows Server 2022, Linux and macOS ARM64, not just compilation.

Downloaded Linux C03/D artifact 13034953549, Windows Server artifact 13034928618, macOS ARM64 artifact 13035218221 and independently checked all 15 C03 observation TOML records: original source-head SHA, console+TeX-log+PDF byte SHA256, positive PDF header/EOF and pdfinfo one-page result; accepted full-suite audit.json with 5/5 records and zero errors on each OS. Positive packages: core/pgfplots/circuitikz/tikz-cd; Lua-only graphdrawing correctly generates no PDF and specific LuaTeX-required error; no claim of native XeLaTeX compatibility. Each hosted OS reports D smoke 4 PASS + 1 negative PASS + 1 KNOWN_LIMITATION, zero FAIL. Inspected real G08 TeX logs on all three: ./main.tex:3: Undefined control sequence and \\TikzariumIntentionallyUndefinedCommand (the intended macro), source unchanged; this rules out the earlier Windows .\\main.tex incidental false positive for this exact run. D04 #27 still adds permanent fail-closed regression.

Font proof PDF 13034619196 independently decoded: Cyrillic Привет, мир!, text and Unicode math recovered, DejaVu Serif/Sans and Latin Modern Math embedded; PDF SHA256 8ace3d60f540605fc05f61f86ddab38a042a63cf966ca49ad3fd9aff4c420032.

Host CI ONLY. Physical Windows 10 22H2 x64 and M1/M4 GUI+Retina/IME, visual editing, PDF-page interaction and R4 provenance/license acceptance remain NOT_RUN/NOT_ACCEPTED. No user TeX source changed. Ready for lease-checked integration; old #10 must be closed as superseded after merge, not force-merged.