linter: intermittent errors when using jsPlugins + jsdoc-js/no-blank-block-descriptions

[camc314]

A user is reporting an inconsistent number of diagnostics being reported when using oxlint + jsPlugins + the jsdoc-js/no-blank-block-descriptions rule.

thread for context:

https://discord.com/channels/1079625926024900739/1080712072012238858/1516280130916323328

[camc314]

I think I found the root cause. This is similar in shape to #19986, but it is not the RuleContext cache this time.

eslint-plugin-jsdoc’s no-blank-block-descriptions goes through iterateAllJsdocs, which keeps a Set of already-tracked JSDoc comment objects:

• https://github.com/gajus/eslint-plugin-jsdoc/blob/b9934253764c228e0bffd7fbcc1ec880237f3ecb/src/rules/noBlankBlockDescriptions.js#L59
• https://github.com/gajus/eslint-plugin-jsdoc/blob/b9934253764c228e0bffd7fbcc1ec880237f3ecb/src/iterateJsdoc.js#L2175-L2176
• https://github.com/gajus/eslint-plugin-jsdoc/blob/b9934253764c228e0bffd7fbcc1ec880237f3ecb/src/iterateJsdoc.js#L2335-L2351

In ESLint, comment objects are fresh per file. In Oxlint’s JS plugin bridge, comment objects were being pooled and reused across files. That means a JSDoc comment in a later file could have the same JS object identity as a previously tracked comment from another file, so the plugin could incorrectly skip it. The observed thread-count dependence makes sense because different thread counts change the order files reach the JS plugin bridge.

[camc314]

@overlookmotel assigning this one to you to take care of 🙂

[overlookmotel]

Have submitted a PR in eslint-plugin-jsdoc to fix: gajus/eslint-plugin-jsdoc#1706

Luckily, the root cause also creates a memory leak in ESLint, so we don't need to rely on "please change this just for Oxlint".

[RioShopfront]

I have tested this issue after upgrading to the version 63.0.6 of eslint-plugin-jsdoc and it seems to be working well in oxlint 1.70 👍

1, 2, 6 and 12 threads all reliably/routinely produce errors for no-blank-block-descriptions