[matt-edmondson]

Fixes #193

What was wrong

DoesMatchRegex returned false for text with no words ("", " ") before it tried the ByWordAny whole-text match added for #125. As a result, a regex written to find blank entries (^\s*$, \s+, ^$) missed exactly those entries under the default ByWordAny, even though it matched them under ByWholeString.

Change

The ByWordAny whole-text attempt now runs before the no-words return. That return stays in place for ByWordAll, so blank text still can't match vacuously there (#107/#109).

Tests

• RegexByWordAnyMatchesBlankTextWithAPatternForBlankText: six blank or whitespace-only texts against blank-finding patterns, under ByWordAny and ByWholeString.
• RegexByWordAllStillMatchesNoBlankText: the same kind of patterns still match nothing under ByWordAll.
• RegexByWordAnyDoesNotMatchBlankTextWithAPatternThatNeedsAWord: zzz and \w still don't match blank text.
• RegexByWordAnyFilterKeepsTheBlankEntries: Filter with ^\s*$ keeps "" and " " and drops the rest.
• With the TextFilter.cs change reverted, 7 of the new cases fail. The ByWordAll and needs-a-word cases pass either way and stay in as guards.
• Full suite on Linux / .NET 10: 279/280. The one failure, EveryMethodInTheReadmeApiReferenceExistsWithItsDocumentedDefaults, fails the same way on main at 5f420d9 and is addressed by Hold only the visible overloads to the README's documented defaults [patch] #186.

---

Generated by Claude Code

[matt-edmondson]

ci / .NET / Test on ubuntu-latest failed on EveryMethodInTheReadmeApiReferenceExistsWithItsDocumentedDefaults. This PR didn't cause it. The same test fails on main at 5f420d9 (CI runs 96, 104 and 107 are all red), and this PR doesn't touch the README or its test.

The fix already exists in #186, and it is test-only, so 28b8ef4 ports it here verbatim. Whichever of the two PRs lands first, the other merges cleanly. Locally, 280/280 tests now pass.

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate failed

Failed conditions
D Maintainability Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE

[matt-edmondson]

All tests on Windows, Linux and macOS, plus Analyze & Release, are green at 28b8ef4. The only red check is SonarCloud Code Analysis (D maintainability rating on new code), and this PR didn't cause it:

• SonarCloud lists 12 "new code" issues, and none of them is on a line this PR changes. The four S3427 blockers are the TextFilter.BinaryCompatibility.cs overload overlaps. The S3776 is ExpandGlobClasses (TextFilter.cs:512). The MSTEST0037 and MSTEST0068 infos are at TextFilterTests.cs 1172, 1370–1392, 1412 and 1603. This PR changes the body of DoesMatchRegex, inserts tests at TextFilterTests.cs 1208–1248, and changes ReadmeTests.cs.
• The likely cause is that main has no analysis that passed its tests: CI on main has been red at 5f420d9 since the README test broke. Without that baseline, SonarCloud counts existing code as new. The other open TextFilter PRs Count only doubled **/ against the zero-segment expansion cap [patch] #187, Keep case-sensitive globs off DotNet.Glob's global default [patch] #189 and Match the empty remainder with a trailing whole-segment ** [patch] #192 fail the same gate in the same way.
• Nothing in this PR can fix it. Once Hold only the visible overloads to the README's documented defaults [patch] #186 (or this PR) turns main green, a fresh main analysis should reset the new-code baseline. After that, a re-analysis of this PR should pass.

---

Generated by Claude Code