Repository navigation
Tokenize yarn-classic .yarnrc lines in one formats/yarn helper: the crawler's copy keeps single quotes that redirect's reader strips #1901
Description
[agent] Filed by the scheduled architecture audit (audit-core). Register: register comment.
Kind: refactor. Register ID: A42.
Problem
Yarn classic's .yarnrc (one key value pair per line) is tokenized twice:
crawlers/npm_crawler.rsparse_yarnrc_modules_folderandsplit_yarnrc_token, L344-L389.`` A double-quoted string has its\escapes removed by hand, and any other value runs to the next whitespace.patch/redirect/yarnrc.rsyarnrc_settingandyarnrc_token, L140-L168 and L447-L490.```` This copy follows yarn's lockfile tokenizer: a double-quoted string is decoded as JSON, a single-quoted string is accepted (yarn's YAML fallback), a bare word ends at a space,:or `,`, and a token yarn can't read is `None`.
The two have drifted. Given --modules-folder 'deps', the crawler resolves the folder 'deps' with the quotes kept, while redirect reads deps. Given --modules-folder ~/x, the crawler accepts the value, while redirect rejects it as unreadable. (This is from reading the code; it was not executed.) Neither tokenizer belongs where it is, because formats/yarn/rc.rs holds only the berry YAML reader.
Proposed change
Move one line reader into formats/yarn/ (for example formats/yarn/classic_rc.rs), based on redirect's yarnrc_token since it is the one that matches yarn. It would offer classic_rc_value(text, key) -> Option<String> (last setting wins) and the tokenizer. Make npm_crawler::yarnrc_modules_folder and redirect/yarnrc.rs callers of it, and delete split_yarnrc_token and the crawler's private line loop.
Size: 3 files, about −70/+30 lines.
Acceptance criteria
- Existing
.yarnrctests pass:redirect/yarnrc_tests.rs, and thenpm_crawlermodules-folder tests. - A new test pins a single-quoted
--modules-folder. - A new test pins a value yarn can't tokenize: it is ignored by both callers.
Activity
mikolalysenko commentedon Nov 1, 2026
[agent] Triage: priority:p1 (yarn classic). No duplicate or open PR found. It touches
yarnrc_modules_folderlike #1902, but the cause is separate: two drifted tokenizers, not the relative--cwd. One fix won't close both.Generated by Claude Code
mikolalysenko commentedon Nov 1, 2026
[agent] Claiming this issue (shared root cause:
.yarnrclines tokenized by two drifted readers). Branch: agent/fix-yarnrc-classic-tokenizer. Claim-ID: 2026-11-01T06:21:14Z-e4a8f6Generated by Claude Code
mikolalysenko commentedon Nov 1, 2026
[agent] Draft PR: #1915
Generated by Claude Code
Read .yarnrc lines through one yarn tokenizer
Read single-quoted .yarnrc values like yarn 1
Read .yarnrc as yarn 1 reads the whole file
Match js-yaml and yarn's tokenizer in .yarnrc
mikolalysenko commentedon Nov 1, 2026
[agent] The audit-core run of 2026-11-01 found that the yarn-1 token grammar has two more copies inside
formats::yarn, beyond the two in this issue.The extra copies:
split_key_patternstoggles on every"and ignores escapes. Executed:"a\"b@1"comes back asa\b@1.yarn_tokenkeeps\", and a bare word there ends only at whitespace. Its doc (L196-197) claims it agrees withsplit_key_patterns, and it doesn't.redirect/yarnrc::yarnrc_tokenJSON-decodes, and a bare word there ends at:, space or,.This is not a request to widen PR #1915. If the merged helper lands in
formats/yarn/patterns.rs, these two can follow it in a later PR.Generated by Claude Code
Cap .yarnrc block nesting before recursing
Drop unused strip_bom import in npm crawler