[webpatser]

DocumentRoot::loadMimeFileTypes() truncates any subtype that contains a dot or a plus sign after the first segment. The regex only allows [a-z0-9\-]+ with one optional +suffix, so with the bundled resources/mime:

Extension	Expected	Served today
xls	application/vnd.ms-excel	application/vnd
docx	application/vnd.openxmlformats-officedocument.wordprocessingml.document	application/vnd

Every vnd.* type in the bundled file ends up as application/vnd, which makes browsers fall back to generic handling for those downloads.

This came out of the fixes I made in my fork (webpatser/fledge-fiber), sending them back upstream as asked. Hope it catches a clean wave.

Summary

• Subtype pattern is now [a-z0-9][a-z0-9.+\-]*, so dotted, plus and dash subtypes are kept whole.
• No change to the mime file itself.

Test plan

• testBundledMimeFileIsParsedInFull: every line of the bundled mime file maps to exactly its listed type (last entry wins for duplicate extensions, matching the parser).
• testDottedPlusAndDashSubtypesAreNotTruncated: xls, svg and docx resolve to their full types.
• Both tests fail on 2.x without the fix and pass with it; full suite green.

[webpatser]

Heads up on the red CI: the jobs fail at "Set up job" because the workflow still uses actions/cache@v2, which GitHub now refuses to run. The 2.x branch fails the same way, so it's unrelated to this change. Locally the full suite is green (2051 tests), and the two new tests fail without the fix.

Happy to send a separate PR bumping the cache action to v4 if that helps.