[matt-edmondson]

Fixes #159

What was wrong

DotNet.Glob reads a [ that has no closing ] as a character class running to the end of the token. ExpandGlobClasses returned any token without a ] unchanged, so a literal bracket silently became a class:

Filter	Before	After
log[1	log1	log[1
a[b	ab	a[b
-log[1	hid log1, kept log[1	hides log[1, keeps log1
log[, a[[]b	literal (unchanged)	literal (unchanged)

Change

ExpandGlobClasses now writes an unclosed [ as the literal [[], the way fnmatch and bash treat it. This handles several brackets in one token and an unclosed bracket after a closed class ([ab]x[c → [ab]x[[]c).

The file[0- question: the issue asked for an explicit decision. I kept the documented #106/#110 behaviour: a class left open after a dash is a range still being typed, so it stays unparseable and is ignored. Two reasons:

• Making it literal would change documented behaviour and its existing tests.
• Open PR Drop an unparseable optional glob token instead of letting it match everything [patch] #162 (for A half-typed second glob token makes every item match under ByWordAny: "*.cs file[0-" shows readme.md and logo.png #141) already changes how an unparseable optional token is treated.

The DoesMatchGlob remarks now describe both cases.

Tests

• New GlobWithAnUnclosedBracketMatchesTheBracketLiterally uses the issue's repro items. It checks IsMatch("a[b", "a[b") is true and IsMatch("ab", "a[b") is false, that -log[1 excludes log[1 and keeps log1, and that the escaped form is unchanged. Each check runs case-sensitive and case-insensitive.
• Four new ExpandGlobClassesRewritesOnlyClassesDotNetGlobMisreads rows, plus one pinning that a[[]b is unchanged.
• With the source change reverted, 5 tests fail. With it, all 207 pass.
• The existing half-typed-range tests (file[0-, [!a-, -file[0-, …) still pass unchanged.

Open PRs #162, #163 and #164 edit the same class rewriter, so whichever lands second may need a rebase.

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
4 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

[matt-edmondson]

ci / .NET / Test on * is red on 34f1bbe. The failure comes from main, not from this PR.

The failing test is ReadmeTests.EveryMethodInTheReadmeApiReferenceExistsWithItsDocumentedDefaults. It fails the same way on main at 5f420d9, on all three OSes. Two PRs that merged close together conflict:

• Restore the pre-1.7 IsMatch, Filter and DoesMatchGlob signatures for older callers [patch] #166 added the [EditorBrowsable(Never)] binary-compatibility overloads in TextFilter.BinaryCompatibility.cs. By design, their filterType and textFilterMatchOptions parameters have no defaults.
• Use the real API names in the README samples and reference [patch] #178's test requires every overload of a documented method to carry the documented defaults. The hidden overloads therefore fail it with IsMatch(filterType) is documented as defaulting to TextFilterType.Glob and three similar messages.

No fix exists on main yet. This patch makes the test ignore the hidden compatibility overloads. With it, all 274 tests pass locally:

--- a/TextFilter.Test/ReadmeTests.cs
+++ b/TextFilter.Test/ReadmeTests.cs
@@ using System.Collections.Generic;
+using System.ComponentModel;
 using System.IO;
@@ public void EveryMethodInTheReadmeApiReferenceExistsWithItsDocumentedDefaults()
-			MethodInfo[] overloads = [.. methods.Where(m => m.Name == methodName)];
+			MethodInfo[] overloads = [.. methods.Where(m => m.Name == methodName && m.GetCustomAttribute<EditorBrowsableAttribute>()?.State != EditorBrowsableState.Never)];

I haven't added it to this PR, to keep this PR to #159. Once it lands on main, merging main in here should turn this PR green.

---

Generated by Claude Code