Repository navigation
fix: align Unicode diagrams and preserve relationship geometry - #53
Conversation
leboiko
left a comment
There was a problem hiding this comment.
GPT-6.1 independent runtime, security/performance and tests/reliability passes found seven renderer/test findings below plus one existing integration-harness reliability gap. All are being resolved in this PR; no follow-up issues.
Major — tests/check_links.rs:16, tests/export_html.rs:14 and tests/section_extract.rs:15 read CARGO_BIN_EXE_markdown-reader at runtime, though Cargo supplies it at compile time. Rust1.92 full tests with a custom target directory fail with OS NotFound because the fallback targets the worktree target/debug path. Use the compile-time Cargo binary path and rerun full suites without a binary-path override.
WezTerm/WSL GUI is unavailable; local raster/protocol tests cannot establish the original visual symptom is fixed. This limitation remains explicit.
leboiko
left a comment
There was a problem hiding this comment.
Additional Major correctness finding from final 6.1 runtime re-review (same PR, no follow-up): sequence render skips a message with an unknown participant, but message_arrow_rows uses push rather than the original message index. Prepending a malformed message and shifting valid activation indices deletes all eight existing bar cells. Store arrow rows by original message index (optional for invalid messages), and process note/event boundaries for skipped messages. A strong failing library-API regression pins the exact valid-message/activation geometry.
|
All findings addressed in 8697672 with GPT-6.1 fix-review-findings.
Independent final runtime and tests re-reviews found no unresolved findings. All local gates exit 0: format; workspace/all-target clippy with warnings denied; full workspace suites on stable and Rust1.92; deny; audit; default docs (existing link warnings). Reports and failing-first evidence are committed in PR-53-review.md and scope notes. No new follow-up issue was created. #48's reported WezTerm/WSL GUI remains unavailable, so the PR ships verified fallback/protocol/raster behavior and keeps that existing issue open for external visual verification. This stack depends on #51; retarget to master after its merge for the workflow's master-only PR trigger. No release/publish was triggered. |
leboiko
left a comment
There was a problem hiding this comment.
Final GPT-6.1 re-review: all nine posted findings are resolved in 8697672. Independent runtime/correctness and tests/reliability passes found no unresolved actionable findings. All local integrated gates pass, including complete stable and Rust1.92 workspace suites without binary overrides, fmt/clippy/deny/audit/default docs. Exact repros and resolutions are committed in PR-53-review.md. Residual limit: actual WezTerm/WSL GUI rendering is unverified; #48 remains open and the protocol change is described as a mitigation. Stacked on #51; GitHub CI master-only trigger does not run for this base.
|
Pre-merge verification on 2026-10-13: both exact reviewed commits passed local stable/Rust1.92 complete workspace suites. The combined commit also passed fmt, workspace/all-target clippy with warnings denied, deny, audit and default docs. Original reviews/findings remain resolved; no code changes since final 6.1 review. #51 is merged as 75616be after all ten Rust CI jobs and both Linux/macOS Nix builds/smoke tests passed. #53 is retargeted to master and synchronized at ef2102d; its tree is exactly #48 remains open for the unavailable WSL/WezTerm GUI verification. No release/publish is being triggered. |
|
Final pre-merge checks all passed on ef2102d, whose file tree exactly matches the reviewed/local-tested commit8697672.
All prior 6.1 review findings remain resolved; source/fixtures have not changed. Merge preserves original authored commits. Issues #41/#52 will close with explanatory messages; #48 stays open for unavailable WSL/WezTerm GUI verification. No release/publish was triggered. |
CJK/fullwidth labels keep contiguous text and aligned borders across shared grids, sequence, class/ER boxes and quadrant charts (#52). Left state notes retain a connected route, and nested activation lanes keep message labels clear (#41), including self calls and autonumber prefixes. Sequence notes preserve their anchors and text as opaque annotations; exposed activation spans remain continuous.
WezTerm sessions prefer iTerm2 after a successful terminal query, including WSL with only TERM_PROGRAM detection. Unicode fallback displays equation tags as
(1)instead oftag1(#48). The reported WezTerm/WSL GUI is unavailable: raster ink, cell geometry and protocol selection are verified locally, but that displayed-image symptom still needs external verification, so the existing issue remains open for that terminal verification. No follow-up issue was created.Built on contributor PR #51, now merged into master with @worldnine’s original commit and authorship intact. This PR targets master. Synchronizing master into the branch produced the exact same file tree as the locally validated commit; no source or fixture changed.
Three independent GPT-6.1 review-pr passes and fix-review-findings resolved all nine findings in this PR. Exact failing-first regressions cover continuation ownership, canvas boundaries, note rectangles, activation indices, numbered labels and visible tag ink. Nine related note/activation/autonumber snapshots changed; zero unrelated reclassification. Findings and evidence are committed in
PR-53-review.mdanddocs/scope-*.md.Local macOS validation: format, workspace/all-target clippy with warnings denied, complete workspace suites on stable and Rust1.92 without binary-path overrides, deny, audit and default docs all pass. Existing rustdoc link warnings remain. All ten Rust CI jobs and both Linux/macOS Nix builds and binary smoke tests passed against master on the synchronized branch. Versions: app1.35.3 / mermaid-text0.57.4. No release or publish was triggered.
Closes #52
Closes #41
Related #48