[i-shutov]

C/N2 source disconnect — complete library connect/disconnect source contract

Base main 97c4b44eeee7ac84ba85f789269c3e15a9f528e4 (#49 C/N2 merged). Old #39 not transplanted. Independent C-owned code on this branch, no A/B/D/E/F files edited. Additive types only; B #50 informed before new API code.

• project_disconnect::plan_library_disconnect proposes removal of arrows.meta / positioning / calc / fit from a confirmed literal \\usetikzlibrary preamble. Standalone declaration line or one comma-list token, exact byte-range SourcePatch, source revision/anchor checks, inverse patch for Undo.
• Duplicate imports, inline declaration comments, unknown macro/conditional imports, mixed line endings, unsupported libraries or stale range/revision refuse with reason, leaving the entire user source intact.
• Nine independent Rust tests cover CRLF/Cyrillic/source authority, four libraries, comma-list first/middle/last, preserved whitespace, duplicates, comments, UNKNOWN and atomic conflict/Undo. Existing six generators and Russian help unchanged.
• Docs/contract in extensions/primitives/PROJECT_IMPORTS.md, plan and QA ledger added. No hidden package installation, TeX parser/CST, CAD solver or Lua fallback. The existing five status dimensions remain independent.

Evidence gates

cargo is absent in this sandbox. Rust fmt/check/test/Clippy and existing real XeLaTeX workflows on this exact branch HEAD are PENDING, not PASS. No claim of actual GUI disconnect; B owns its visible diff/confirmation/undo. Hosted CI != physical Windows 10 x64/MiKTeX or Mac M1/M4. External XeLaTeX remains authoritative.

Related: merged #49, B #50, D #40. See plans/2026-10-31_C_N2_LIBRARY_DISCONNECT.md and insights/2026-10-31_C_N2_LIBRARY_DISCONNECT_QA.md.

N1/B10 integration fix on same C PR (2026-10-31)

B #50 identified a real N1 app→XeLaTeX authoring regression: existing C scanner returned UNKNOWN for all imports when the source contained normal flat fontspec selections and a later user macro. C now preserves CONFIRMED source-text evidence from a trusted literal preamble prefix, accepts only simple import-neutral \\setmainfont/\\setmathfont assignments, and keeps absent/macro-dependent or conditional declarations UNKNOWN. Existing connect/disconnect patches remain fail-closed for any opaque preamble; no B/A/D/E/F code changed. LF/CRLF source, all six tool statuses, unsafe conditional/macro cases, plan refusal tests included; updated C schema docs/plan/insight in this same PR. B notified via #50 comment.

Current exact C HEAD: 3d64c60815f7980df67394ce6fd2541f80b75c43. Earlier 38febde0 FAILED rustfmt on new test formatting; corrected with exact hosted diff. Exact-head hosted CI VERIFIED: N1 real app authoring state→generated source→external XeLaTeX SUCCESS (source/PDF artifact n1-app-authoring-linux, id 13043022726); D QA matrix SUCCESS, 7/7 jobs, including Linux/macOS/Windows Server Rust and independent XeLaTeX; native CI SUCCESS, 3/3 jobs, including hosted macOS and Windows Server native Rust; figure-font glyph/embedding gate SUCCESS. These are GitHub-hosted integration results, NOT physical Windows 10 with user-installed MiKTeX, Mac M1/M4 GUI/Retina/IME, or B-owned visible disconnect UI. PR remains draft. Physical Windows 10/MiKTeX and Mac M1/M4, and B GUI disconnect action remain NOT_RUN.

[i-shutov]

Exact-head QA update (2026-10-31): first branch head 5cf5886a35eb4ee4aa8660ae7ce3d5e8205836bb failed Windows Server Rust cargo fmt --check, actual job 124526566951 in CI run 41197516763. The complete rustfmt diff contained precisely two changes (multiline literal_names signature and multiline inline-comment test array). Both fixed without disabling tests. New current head 26a564af24c0363d2bdd100583205710d9acf961: native CI, D QA, N1→XeLaTeX, font gate are still QUEUED/RUNNING, not PASS. There is no local Rust toolchain or physical Win10/Mac QA here. PR remains draft.

[i-shutov]

Evidence on current exact HEAD 26a564af24c0363d2bdd100583205710d9acf961 (2026-10-31): D QA matrix run 41197694448 has completed Rust fmt/tests/Clippy successfully on ubuntu-24.04, macos-15 and windows-2022 (3/3 Rust jobs). The Ubuntu Rust job 124527091142 explicitly logs tests/project_disconnect.rs: 9 passed; 0 failed, original tests/project_imports.rs: 12 passed, primitives 6 passed, source-integrity 17 passed, plus Clippy -D warnings clean. Ubuntu and hosted macOS actual XeLaTeX CLI jobs succeeded; Windows Server TeX job still RUNNING. N1 authoring→XeLaTeX still RUNNING; figure fonts PASS; native CI not yet complete. Do not promote partial matrix to full CI PASS; physical Win10+MiKTeX, real M1/M4 and B's visible disconnect UI remain NOT_RUN. No code/QA gate relaxed.

[i-shutov]

Latest exact-HEAD (26a564af24c0363d2bdd100583205710d9acf961) CI evidence: D QA matrix 41197694448 now shows 7/7 jobs SUCCESS, including all three hosted Rust fmt/tests/Clippy hosts, all three external XeLaTeX CLI hosts, and D05 independent uploaded-artifact evidence. Ubuntu Rust log explicitly proves project_disconnect: 9/9, project_imports: 12/12, primitives: 6/6. N1 app→XeLaTeX 41197694417 PASS and fonts 41197694386 PASS. Native CI 41197694365: Linux XeLaTeX fixture PASS; Windows Server/macOS Rust jobs still RUNNING, so full four-workflow CI PASS not claimed. Physical Win10 MiKTeX/M1/M4 and B UI disconnect NOT_RUN.

[i-shutov]

Integrator gate after merging A/W06 #38 into main 013720ee6f0e1dc29f487124499ed9469d86b089: B10 #50 exact-head ba564771 has a real N1 authoring→XeLaTeX FAIL (run 41197494077, job 124526498804) while its 3-host Rust tests PASS. n1-authoring-proof refuses the first insertion: Подключение пакета/библиотеки не подтверждено. Exact fixture already contains literal \usepackage{tikz} and \usetikzlibrary{arrows.meta}, but project_imports::inspect classifies unrelated \setmainfont{DejaVu Serif}, \setmathfont{latinmodern-math.otf}, \newcommand{\UserPath}{...} as opaque and ProjectImports::status() converts even confirmed literal declarations to UNKNOWN. Please address in C-owned source scanner/contract, not a duplicate B parser: preserve UNKNOWN for truly ambiguous/conditional/redefined imports and UNKNOWN/Refused for unsafe insertion, but do not make safe standalone known imports unusable merely due unrelated figure-font/user-macro lines. Add a regression with these exact fontspec+macro lines and test arrows.meta confirmed plus N1 proof. Existing #51 source removal tests are green; no merge of #50 until the real N1 proof passes on reconciled source. No changes to your branch by integrator.

[i-shutov]

C/N2 integration update 38febde0c05ad093a339b09ab91236379d656974: fixed actual B10/N1 regression of literal TikZ imports globally UNKNOWN due to later user macro/font declarations. Tests now cover N1 fontspec/unicode-math source with all six tools (LF/CRLF), conditional/unsafe-negative cases. Updated same C PR rather than creating a micro-PR. Original #51 disconnect contract unchanged; pending exact-head hosted Rust/N1 QA. See B #50 for handoff.

[i-shutov]

Final exact-head QA evidence for C/N2 PR #51, 3d64c60815f7980df67394ce6fd2541f80b75c43 (code+tests+schema docs+plans/insights in this same C PR). GitHub Actions now ALL completed SUCCESS: native CI 41202090696 3/3 jobs (hosted macOS, hosted Windows Server, Linux real XeLaTeX); D QA 41202090692 7/7 (Rust fmt/check/test/Clippy plus real external XeLaTeX); N1 authoring proof 41202090704 SUCCESS for the exact former failing fontspec+opaque macro scenario, with generated .tex/PDF and glyph verification, artifact 13043022726; figure-font gate 41202090693 SUCCESS. The initial 38febde0 rustfmt FAIL is preserved in insights as failure history, not rewritten to PASS. Source-level library CONFIRMED remains distinct from INSTALLED, ENGINE_COMPATIBLE, TOOL_IMPLEMENTED, HELP_AVAILABLE; plan_import and plan_library_disconnect stay refusal-only on uncertainty. No GUI implementation claim or physical Win10/MiKTeX / Mac M1/M4 acceptance. B #50 and D #52 have handoff comments; E/F files untouched. Leave draft pending B visible disconnect UI and physical gates.