Fix embedded language classification inside multi-line string #78588
Conversation
| // two"; | ||
| // The check below ensures we correctly return the spans in the correct order, i.e. 'one', '""', 'two'. | ||
| while (spanIndex + 1 < originalClassifiedSpans.Count && | ||
| textSpan.Contains(originalClassifiedSpans[spanIndex + 1].TextSpan)) |
There was a problem hiding this comment.
this was buggy when there was a multi-line classification that had another multi-line classification inside of it. For example, in
// lang=c#-test
const string Code = @"
class C {
//
}";
the entire (multi-line) string is classified as roslyn-test code, but there are also verbatim string classifications for all the whitespace in between, which also spans multiple lines (e.g from after the { to before the //).
As we create new spans line by line for roslyn-test classification, this code tries to insert classifications that are on the same line into the set, before it proceeds to the next line (to keep everything ordered). But that only happens for single line spans. If there is a multi-line span inside the one we're breaking up, this exits and does not add it until after the entire original multi-line is added, causing the spans to be out of order.
Instead of having this complexity, we can just break up the multi-line spans and sort the set afterwards.
3 participants
Resolves #74809