Skip to content

91_mds037-skip-generated-sections.md

Latest commit

 

History

History
83 lines (72 loc) · 2.94 KB

91_mds037-skip-generated-sections.md

File metadata and controls

83 lines (72 loc) · 2.94 KB
Raw
id 91
title MDS037 skips paragraphs inside generated sections
status ✅
summary Paragraphs inside `<?include?>` and `<?catalog?>` directive bodies are copies of content owned by another file; MDS037 must not count them as cross-file duplicates.

MDS037 skips paragraphs inside generated sections

Builds on redundancy detection work to skip paragraphs inside generated sections (MDS037 duplicated-content).

Goal

Stop MDS037 from flagging paragraphs produced by the include or catalog directive. The rule then runs cleanly on a project that shares prose across agent instruction files or renders catalog tables from front matter.

Context

MDS037 walks paragraphs via ast.KindParagraph and does not look at the surrounding processing instructions. Two directives break that:

  • <?include file: X ?> expands prose from X into the host file. After expansion the host file and X hold the same paragraph text on disk, and MDS037 reports each paragraph as duplicated. A single project often points two or three agent files at the same shared source; every run fires.
  • <?catalog glob: "..." ?> renders a table or list whose row template pulls from each target's front matter (e.g. {summary}). When a target also states the same summary as a sentence in its body, the catalog row duplicates that sentence. The catalog block is a table today, so the line usually sits inside a table cell, but a row template can produce paragraph-shaped output too.

The ranges are easy to find. mdsmith parses the open and close tags as *lint.ProcessingInstruction siblings. The rule can map those to byte ranges without re-parsing the file.

Tasks

  1. Walk top-level AST once to build a list of generated-section byte ranges. A range starts at the opening PI's segment end and closes at the closing PI's segment start.
  2. In extractParagraphs, skip any paragraph whose first-line byte offset sits inside one of those ranges.
  3. Likewise skip the paragraph when the rule walks other files during corpus indexing — otherwise a host file would still match against source text the other file only carries because of its own generated section.
  4. Update the README to document this behavior and drop the recommendation to hand-exclude the host files.
  5. Add tests: include-expanded paragraph in host is not flagged; catalog-rendered paragraph in host is not flagged; duplicates outside generated sections still fire.

Acceptance Criteria

  • A paragraph in a file's body whose bytes sit between <?include?> / <?/include?> markers does not produce an MDS037 diagnostic.
  • Same for <?catalog?> / <?/catalog?>.
  • A real duplicate outside any generated section still fires.
  • mdsmith check . stays clean on this repo with MDS037 enabled in .mdsmith.yml.
  • All tests pass: go test ./...
  • go tool golangci-lint run reports no issues