Stop a half-typed excluded glob token hiding every item

TextFilter.Test/TextFilterTests.cs

@@ -779,11 +779,14 @@
 	[DataRow("[!a-")]
 	[DataRow("*[0-")]
 	[DataRow("[a-]")]
+	[DataRow("-file[0-")]
+	[DataRow("!file[0-")]
+	[DataRow("^file[0-")]
 	public void GlobWithAHalfTypedRangeDoesNotThrow(string pattern)
 	{
 		// DotNet.Glob's tokeniser throws IndexOutOfRangeException on a range left open after the dash,
 		// which every type-ahead filter passes through while the user types "file[0-9].txt". An
-		// unparseable token degrades to match-anything, as an invalid regex already does.
+		// unparseable token is ignored: plain and required tokens match anything, excluded ones exclude nothing.
 		foreach (TextFilterCaseSensitivity caseSensitivity in new[] { TextFilterCaseSensitivity.CaseSensitive, TextFilterCaseSensitivity.CaseInsensitive })
 		{
 			Assert.IsTrue(TextFilter.IsMatch("file1.txt", pattern, TextFilterType.Glob, TextFilterMatchOptions.ByWordAny, caseSensitivity));
@@ -793,6 +796,23 @@
 		}
 	}
 
+	[TestMethod]
+	[DataRow("-readme[a-")]
+	[DataRow("!readme[a-")]
+	[DataRow("^readme[a-")]
+	public void ExcludedGlobWithAHalfTypedRangeExcludesNothing(string pattern)
+	{
+		// An unparseable excluded token is ignored while the user finishes typing it. Treating it as
+		// match-anything, as unprefixed tokens are, would exclude every item on each keystroke.
+		List<string> items = ["file1.txt", "file2.txt", "readme.md"];
+		foreach (TextFilterCaseSensitivity caseSensitivity in new[] { TextFilterCaseSensitivity.CaseSensitive, TextFilterCaseSensitivity.CaseInsensitive })
+		{
+			CollectionAssert.AreEqual(
+				items,
+				TextFilter.Filter(items, pattern, TextFilterType.Glob, TextFilterMatchOptions.ByWordAny, caseSensitivity).ToList());
+		}
+	}
+
 	[TestMethod]
 	[DataRow("", TextFilterType.Regex)]
 	[DataRow("   ", TextFilterType.Regex)]

TextFilter/TextFilter.cs

@@ -82,7 +82,8 @@
 	private static HashSet<char> ExcludedTokenPrefixes { get; } = ['!', '-', '^'];
 	private static HashSet<char> RequiredTokenPrefixes { get; } = ['+'];
 	private static ConcurrentDictionary<string, Regex> RegexCache { get; } = [];
-	private static ConcurrentDictionary<string, Glob> GlobCache { get; } = [];
+	// A null entry records a token that could not be parsed, so the parse is not retried on every keystroke.
+	private static ConcurrentDictionary<string, Glob?> GlobCache { get; } = [];
 
 	// Filter patterns are caller-supplied, and in the keystroke-driven filter box this library exists
 	// for, every prefix of what the user types becomes its own key. Unbounded, the caches therefore
@@ -123,10 +124,8 @@
 	private static readonly TimeSpan RegexMatchTimeout = TimeSpan.FromSeconds(1);
 
 	[System.Diagnostics.CodeAnalysis.SuppressMessage("Performance", "SYSLIB1045:Convert to 'GeneratedRegexAttribute'.", Justification = "Not available in older frameworks")]
 	private static Regex RegexMatchAnything() => new(".*", RegexOptions.Compiled);
 
-	private static Glob GlobMatchAnything() => Glob.Parse("*");
-
 	/// <summary>
 	/// Gets a hint for the specified filter type.
 	/// </summary>
@@ -277,7 +276,7 @@
 
 			return ExcludedTokenPrefixes.Contains(prefix)
 				? TextFilterTokenType.Excluded
 				: RequiredTokenPrefixes.Contains(prefix)
 				? TextFilterTokenType.Required
 				: TextFilterTokenType.Optional;
 		})
@@ -303,8 +302,9 @@
 	/// <param name="caseSensitivity">Whether the match distinguishes uppercase from lowercase.</param>
 	/// <returns><c>true</c> if the text matches the glob filter pattern; otherwise, <c>false</c>.</returns>
 	/// <remarks>
-	/// A token that cannot be parsed as a glob, such as the half-typed range <c>file[0-</c>, matches
-	/// everything rather than throwing, the same way an invalid regex pattern does.
+	/// A token that cannot be parsed as a glob, such as the half-typed range <c>file[0-</c>, is ignored
+	/// rather than throwing: as a plain or required token it matches everything, and as an excluded
+	/// token it excludes nothing.
 	/// </remarks>
 	public static bool DoesMatchGlob(string text, string filter, TextFilterMatchOptions textFilterMatchOptions, TextFilterCaseSensitivity caseSensitivity = TextFilterCaseSensitivity.CaseSensitive)
 	{
@@ -334,7 +334,9 @@
 			optionalTokens = [];
 		}
 
-		bool anyExcludedMatches = excludedTokens.Any(filterToken => AnyTokenMatchesGlobFilter(filterToken, textTokens, caseSensitivity));
+		// An unparseable excluded token is skipped rather than treated as match-anything, which here
+		// would exclude every item while the user is still typing the token.
+		bool anyExcludedMatches = excludedTokens.Any(filterToken => ResolveGlob(filterToken, caseSensitivity) is Glob glob && textTokens.Any(glob.IsMatch));
 
 		if (anyExcludedMatches)
 		{
@@ -378,9 +380,9 @@
 		Ensure.NotNull(filterToken);
 		Ensure.NotNull(textTokens);
 
-		Glob glob = ResolveGlob(filterToken, caseSensitivity);
+		Glob? glob = ResolveGlob(filterToken, caseSensitivity);
 
-		return textTokens.Any(glob.IsMatch);
+		return glob is null || textTokens.Any(glob.IsMatch);
 	}
 
 	/// <summary>
@@ -395,12 +397,13 @@
 		Ensure.NotNull(filterToken);
 		Ensure.NotNull(textTokens);
 
-		Glob glob = ResolveGlob(filterToken, caseSensitivity);
+		Glob? glob = ResolveGlob(filterToken, caseSensitivity);
 
-		return textTokens.All(glob.IsMatch);
+		return glob is null || textTokens.All(glob.IsMatch);
 	}
 
-	private static Glob ResolveGlob(string filterToken, TextFilterCaseSensitivity caseSensitivity)
+	// Returns null for a token that cannot be parsed, so each caller can decide what ignoring it means.
+	private static Glob? ResolveGlob(string filterToken, TextFilterCaseSensitivity caseSensitivity)
 	{
 		string cacheKey = CacheKey(filterToken, caseSensitivity);
 
@@ -416,10 +419,9 @@
 			{
 				// DotNet.Glob's tokeniser throws IndexOutOfRangeException on a range left open after the
 				// dash ("file[0-"), which is ordinary intermediate input while someone types a range into
-				// a filter box. Degrade the way an invalid regex does: match anything, and cache that so
-				// the exception is not raised again on every keystroke. Caught broadly so the next
-				// tokeniser bug is contained too.
-				glob = GlobMatchAnything();
+				// a filter box. Cache it as unparseable so the exception is not raised again on every
+				// keystroke. Caught broadly so the next tokeniser bug is contained too.
+				glob = null;
 			}
 
 			AddBounded(GlobCache, cacheKey, glob);