fix(lint): never throw on odd frontmatter or huge entities, and place and find diagnostics correctly by onmax · Pull Request #43 · onmax/comark-plugins

onmax · 2026-10-20T18:16:48Z

Follows #38. Fixes the review findings on comark-lint. Each one has a regression test that goes through lint().

Crash on ---title: x\n---\n::bad. source.ts now applies Comark's frontmatter rule: the YAML starts at the fifth character, even on the first line, and an empty block isn't stripped. Token lines are offset by the closing --- line, the line Comark's remaining content starts on, instead of state.parsedLines, which miscounts this case. The token-to-source lookups are guarded. lint() also runs each rule on its own, so a rule that crashes reports one diagnostic under its own rule id (naming the error and where to report it) and the other rules still run.
Crash on &#99999999999…;. A character reference past U+10FFFF, including one so long it parses as Infinity, decodes to U+FFFD as in a browser. Before, % 0x110000 turned Infinity into NaN.
comark/unclosed-block false positives. Lines in html_block, code_block, and fence tokens are skipped by comark/unclosed-block and comark/malformed-attributes. I couldn't trigger the indented-code case, because the opener pattern already needs three spaces or fewer. The skip covers it anyway, and it also covers fences nested deeper than the source scan sees.
Unknown component misses. mdc_block_shorthand openers count as block components, so :badge[x] at the start of a line, nested or not, gets reported. An inline token without a map takes its block's map, so :chip inside :pill[see :chip[y]] is found too. This also numbers shorthand blocks for comark/plugin-issue placement.
Diagnostics on the first match in the source. Unsafe URLs and broken anchors are now matched only where the value starts an attribute or link, after ", ', =, (, or <. The same text in prose is skipped. I chose this over tracking element line ranges because Comark nodes carry no positions, and mapping elements back to tokens would take far more code. The check also gets two links on the same line right, which line ranges wouldn't.
Duplicate attributes after a nested brace value. The {…} scanner understands nesting and \ escapes inside quotes. It both finds the closing } and splits the attributes, so [x]{foo={"x":"}"} foo=2} reports foo twice. This matches how Comark parses it (it keeps foo=2).

The docs page says how a crashing rule reports itself, that HTML blocks are skipped, and how URLs are placed.

pnpm ready passes: 156 tests, Vite Doctor clean. The two unbound-method warnings in e2e/annotate.e2e.ts are already on main.

🤖 Generated with Claude Code

… and find diagnostics correctly - Take the frontmatter's closing line the way Comark does, so YAML on the `---` line no longer shifts token lines past the source. - Run each rule on its own; a rule that crashes reports one diagnostic and the others still run. - Read character references past U+10FFFF as U+FFFD instead of throwing. - Skip HTML blocks and code in comark/unclosed-block and comark/malformed-attributes. - Report `:name[…]` shorthand blocks and components inside them as unknown components. - Place unsafe URLs and broken anchors where they start an attribute or link, not on the same text in prose. - Read attribute values with nested braces and escaped quotes, so a key set twice after one is still found.

vitehub-bot · 2026-10-20T18:16:56Z

Session	Status	Started	Duration
Pending	Completed	Not started	—

Latest result

Pull request merged.

Final answers

Session

Pull request merged.

Session

Waiting for GitHub merge request 244d772d-0d09-4ab6-af06-f57770db79b5.

Session