Wrong summary for generated regex

[pawchen]

Description

    [GeneratedRegex(@"\b(ifn?def|endif)?")]
    private static partial Regex MyRegex1 { get; }

The summary generated should state it matches the string if before the optional n, but instead it states Match an empty string.:

I examined the generated matching code and it did look for the 'i' then the 'f', so it appears only the summary was wrong.

Reproduction Steps

    [GeneratedRegex(@"\b(ifn?def|endif)?")]
    private static partial Regex MyRegex1 { get; }

Expected behavior

summary should be stated correctly

Actual behavior

summary was wrong

Regression?

No response

Known Workarounds

Don't look at it?

Configuration

VS 17.13.6
.NET 9

Other information

No response

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-text-regularexpressions
See info in area-owners.md if you want to be subscribed.

[steveharter]

@pawchen can you please open an issue here: https://developercommunity.visualstudio.com/VisualStudio?q=regex

[pawchen]

@steveharter I believe the source code for the generated summary is in this repo.

[huoyaoyuan]

Yes, the summary is generated by the Regex source generator. /cc @stephentoub

[stephentoub]

The problem is here:

runtime/src/libraries/System.Text.RegularExpressions/gen/RegexGenerator.Emitter.cs

Lines 1839 to 1849 in 29e3af6

	case RegexNodeKind.Concatenate when child.Child(0) == startingLiteralNode && (startingLiteralNode.Kind is RegexNodeKind.One or RegexNodeKind.Set or RegexNodeKind.Multi):
	// This is a concatenation where its first node is the starting literal we found and that starting literal
	// is one of the nodes above that we know how to handle completely. This is a common
	// enough case that we want to special-case it to avoid duplicating the processing for that character
	// unnecessarily. So, we'll shave off that first node from the concatenation and then handle the remainder.
	// Note that it's critical startingLiteralNode is something we can fully handle above: if it's not,
	// we'll end up losing some of the pattern due to overwriting `remainder`.
	remainder = child;
	child = child.Child(0);
	remainder.ReplaceChild(0, new RegexNode(RegexNodeKind.Empty, remainder.Options));
	goto HandleChild; // reprocess just the first node that was saved; the remainder will then be processed below

The emitter is mutating the regex node tree, for convenience and under the assumption that after this point no one else will be looking at the tree. But the XML comment generator does.