[matt-edmondson]

Fixes #198

What changed

ExtractGlobFilterTokens always treated a token's first character as a prefix. Because of that, an item named +page.svelte couldn't be found by typing its name, and !important.txt or ^notes.md let every item through.

• Escape: a backslash before a leading +, -, ! or ^ makes that character literal.
  • \+page.svelte is the optional token +page.svelte.
  • The escape also works after a prefix: +\+page* is the required token +page*, and -\-draft* is the excluded token -draft*.
  • A backslash anywhere else is still a path separator, so src\main.cs matches as before.
• GetHint(TextFilterType.Glob): now says that ! and ^ also exclude, and explains the escape.
• README: a new "Glob Tokens" section lists the three token kinds and every prefix, with the backslash escape and the [+] class workaround.

Behaviour change

A glob token that starts with \ followed by a prefix character used to mean "a path separator, then that character". It now means the literal character. Matching a leading separator followed by + still works with [\\]+… or /+….

Tests

• Each of the four prefix characters, escaped, finds exactly its item. So do \+page* and [+]page.svelte.
• An escaped prefix after a required prefix and after an excluded prefix.
• Unescaped +, -, ! and ^ still mean required or excluded.
• A backslash before an ordinary character is still a path separator.
• The GetHint expectation is updated.

With the unescape stubbed out, 9 of the new cases fail. With the fix, every new test passes. dotnet test gives 279/280. The one failure, ReadmeTests.EveryMethodInTheReadmeApiReferenceExistsWithItsDocumentedDefaults, also fails on main (5f420d9) without this change.

As the triage noted, open PRs that edit README.md and ReadmeTests.cs may textually conflict with this one. No other open PR changes how prefixes are parsed.

---

Generated by Claude Code

[matt-edmondson]

This PR inherits the ci / .NET / Test failure that's on main (5f420d9), where ReadmeTests.EveryMethodInTheReadmeApiReferenceExistsWithItsDocumentedDefaults already fails. I've cherry-picked the existing fix from open branch fix/183-readme-api-test-compat-overloads (9328e00) as dc50700. With it, dotnet test passes 280/280 locally.

---

Generated by Claude Code

[matt-edmondson]

The SonarCloud gate (Maintainability D on new code) has the same cause as on #199. The only analysis of main is from 2026-11-04 (837b460), because every CI run on main since then has failed on the README test. "New code" therefore includes the four pre-existing S3427 BLOCKERs on the TextFilter.BinaryCompatibility.cs overloads and the S3776 in ExpandGlobClasses, and this PR touches neither. The findings on lines this PR adds are seven INFO-level MSTEST0068 suggestions (CollectionAssert.AreEqual → Assert.AreSequenceEqual) in the new tests. The rest of the suite also uses CollectionAssert.AreEqual, and changing those seven alone would not lift the rating to A.

The gate should clear once #186 merges, main goes green, and SonarCloud re-baselines.

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
7 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud