Skip to content

2606171404_parity-layer1-inline-rules.md

Latest commit

 

History

History
92 lines (77 loc) · 4.21 KB

2606171404_parity-layer1-inline-rules.md

File metadata and controls

92 lines (77 loc) · 4.21 KB
Raw
id 2606171404
title Parity parse-skip: migrate the Layer-1 inline rules
status ✅
summary Add a nil-AST path to the parity rules that read inline content — heading text, links, emphasis, code spans, inline HTML, reference definitions — by driving the shared per-block inline re-parse (lint.InlineBlocks) instead of the full document parse, each gated byte-identical to the AST across the corpus.
model opus
depends-on
2606171258

Parity parse-skip: migrate the Layer-1 inline rules

Goal

Move the inline-content parity rules onto the Layer 1 path, so they read re-parsed inline spans instead of forcing the document parse.

Background

These rules read flattened inline content, not just lines: a heading's text, a link's text and destination, an emphasis run, a code span, an inline HTML tag, or the reference-definition map. The line scan cannot produce that. The Layer 1 index (lint.InlineBlocks) re-parses one block's inline tree on demand, so a rule reaches inline spans without the whole-document parse.

Rule Name Reads
MDS005 no-duplicate-headings heading text
MDS017 no-trailing-punctuation-in-heading heading text
MDS041 no-inline-html inline HTML nodes
MDS042 emphasis-style emphasis markers
MDS049 no-space-in-link-text link text edges
MDS050 proper-names inline text runs
MDS052 no-space-in-code-spans code-span content
MDS053 no-unused-link-definitions ref defs and uses
MDS063 descriptive-link-text link text
MDS068 link-style link form
MDS034 markdown-flavor flavor-specific spans

Tasks

For each rule (all done):

  1. Add the nil-AST path that drives lint.InlineBlocks for the blocks the rule cares about, reusing the existing inline extraction. MDS053 is cross-block: its reference def/use map is assembled by walking every block's re-parsed inline spans.
  2. Keep the diagnostic byte-identical: same line, column, message.
  3. Regenerate the walk audit and sync the embedded rulelayer copy.
  4. Add a TestCheck_NilASTMatchesAST unit test covering the inline edge cases — emphasis at a heading's end, a link inside emphasis, a code span holding bracket text.

Result

All eleven rules now serve the parse-skipped (nil-AST) path from the Layer-1 inline parse. Each is gated byte-identical to the AST. The gate is a TestCheck_NilASTMatchesAST unit test plus the corpus equivalence harness.

Eight resolve to the audit's A-no-skipping (Layer 0) category: MDS005, MDS017, MDS042, MDS049, MDS053, MDS063, MDS068, and MDS034. MDS053 assembles its reference def/use map across every run. MDS034 reads bare URLs from the inline runs via the new flavor.BareURLFindingsInTree. It reads alert blockquotes from the Layer 0 BlockQuote spans via the new flavor.IsAlertMarkerLine and AlertFinding. The dual-parser features still detect from the body.

Three stay B-prose-only. MDS041 reads inline HTML and HTML-block content. MDS050 scans code-block and HTML-block bodies under check-code and check-html. MDS052 reads code-span content.

Each of the three is nil-AST-safe with a working inline path. The audit's code-perturbation probe scrambles the very content these rules read, so they are code-content-sensitive by design. MDS066 reached the same terminal classification in plan 2606171402.

Acceptance Criteria

  • Each rule resolves to Layer 0 / Layer 1: eight are A-no-skipping (Layer 0); MDS041, MDS050, MDS052 are B-prose-only (code-content-sensitive by design, like MDS066) yet nil-AST-safe on the Layer-1 inline path.
  • TestLayer0Gate_CorpusDiagnosticsEquivalence green with them on.
  • go test ./... passes (the pre-existing internal/release PGO / code-signing failures are environment-only and untouched here).