Breaking changes in v3.3.0

[jaydenseric]

With v3.2.12 the glob **/{!(*.d).mts,!(*.d).cts,*.{mjs,cjs,js}} used to select these files:

[
  ".cjs",
  ".cts",
  ".js",
  ".mjs",
  ".mts",
  "a.mjs",
  "ad.cts",
  "ad.mts",
  "a/.cjs",
  "a/.cts",
  "a/.js",
  "a/.mjs",
  "a/.mts",
  "a/a.mjs",
  "a/ad.cts",
  "a/ad.mts",
]

But now with v3.3.0 it only selects these files:

[
  ".cts",
  ".mts",
  "a.mjs",
  "ad.cts",
  "ad.mts",
  "a/.cts",
  "a/.mts",
  "a/a.mjs",
  "a/ad.cts",
  "a/ad.mts",
]

This is breaking this test assertion in find-unused-exports, which uses fast-glob indirectly as a dependency of a dependency (globby):

https://github.com/jaydenseric/find-unused-exports/blob/dd74e679e819c84e606d683f79d9ada1181a20b9/MODULE_GLOB.test.mjs#L15-L41

When using a package.json field overrides I was able to determine that the specific cause for the changed behavior of the glob was the fast-glob update from v3.2.12 to v3.3.0:

  {
    "overrides": {
-     "fast-glob": "3.2.12"
+     "fast-glob": "3.3.0"
    }
  }

It took me a lot of work to figure out the glob originally, and I'm not really sure what I can do in find-unused-exports to workaround this unexpected issue. I can't pin a particular version of fast-glob that still works because it's a dependency of a dependency (globby). It would be fantastic if this could be resolved in a speedy fast-glob patch release, assuming I'm not doing something blatantly incorrect with in my glob I could update to a more correct version.

Environment

• OS Version: macOS v13.4
• Node.js Version: v20.4.0

[mrmlnc]

It looks like both behaviours are incorrect.

[
  ".cjs", // matched by **/*.cjs when the dot option is true
  ".cts", // matched by **/!(*.d).cts when the dot option is true
  ".js", // matched by **/*.js when the dot option is true
  ".mjs", // matched by **/*.mjs when the dot option is true
  ".mts", // matched by **/!(*.d).mts when the dot option is true
  "a.mjs", // matched by **/*.mjs
  "ad.cts", // matched by **/*.cjs
  "ad.mts", // matched by **/!(*.d).mts
  "a/.cjs", // matched by **/*.cjs when the dot option is true
  "a/.cts", // matched by **/!(*.d).cts when the dot option is true
  "a/.js", // matched by **/*.js when the dot option is true
  "a/.mjs", // matched by **/*.mjs when the dot option is true
  "a/.mts", // matched by **/!(*.d).mts when the dot option is true
  "a/a.mjs", // matched by **/*.mjs
  "a/ad.cts", // matched by **/!(*.d).cts
  "a/ad.mts", // matched by **/!(*.d).mts
]

Correct result:

// echo **/{!(*.d).mts,!(*.d).cts,*.{mjs,cjs,js}}

[
  "a.mjs",
  "a/a.mjs",
  "a/ad.cts",
  "a/ad.mts",
  "ad.cts",
  "ad.mts"
]

In the new version we are closer to the correct result, but it is still incorrect.

Probably related issues:

• Glob **/!(*-dbg).@(js) is wrongly translated into RegExp and thus DOES match -dbg.js files micromatch/picomatch#93
• https://github.com/micromatch/picomatch/pull/102/files

Yeap… the root issue is here:

https://github.com/micromatch/picomatch/blob/13efdf0c7d0e07ee30bf78d0079184aa4a0ec7d4/lib/parse.js#L258

Here the rest variable contains .mts and then the parser returns \\.mts for the **/!(*.d).mts.