Improve performance of FAR #73523
Improve performance of FAR #73523
Conversation
dotnet-issue-labeler
bot
added
Area-IDE
untriaged
| public readonly ImmutableDictionary<string, string> AdditionalProperties = additionalProperties; | ||
|
|
||
| public static SerializableSourceReferenceItem Dehydrate(int definitionId, SourceReferenceItem item) | ||
| => new(definitionId, | ||
| SerializableDocumentSpan.Dehydrate(item.SourceSpan), | ||
| // We're always have classified spans for C#/VB, which are the only languages used in OOP find-references. | ||
| SerializableClassifiedSpansAndHighlightSpan.Dehydrate(item.ClassifiedSpans!.Value), |
There was a problem hiding this comment.
the computation always happens here:
roslyn/src/Features/Core/Portable/FindUsages/DefinitionItemFactory.cs
Lines 303 to 310 in cdbed67
we just were dropping the value on the floor here.
| item.SymbolUsageInfo, | ||
| item.AdditionalProperties); | ||
|
|
||
| public async Task<SourceReferenceItem> RehydrateAsync(Solution solution, DefinitionItem definition, CancellationToken cancellationToken) | ||
| => new(definition, | ||
| await SourceSpan.RehydrateAsync(solution, cancellationToken).ConfigureAwait(false), | ||
| // Todo: consider serializing this over. |
There was a problem hiding this comment.
yes. this was a good thing to consider :D
| /// </summary> | ||
| #pragma warning disable IDE0052 // Remove unread private members | ||
| private readonly SemanticModel _nullableEnabledSemanticModel; | ||
| #pragma warning restore IDE0052 // Remove unread private members |
There was a problem hiding this comment.
this is part of the change. ideally nothing will be using nullable-enabled-semantic-model (And i want to stamp out anything that is using it). but if they do, this at least means we cache the results while processing the doc.
There was a problem hiding this comment.
next week i'll be setting BPs in the normal 'get semantic model' codepath so that i can update any final services still using it during FAR to stop.
While computing the results for FAR we were computing classifications for the reference locations. These were getting dropped, only to be recomputed later when trying to present them in the UI. This is, obviously, terrible.
This drops the time doing a FAR for a widespread symbol (like 'SyntaxToken') from 57 seconds on my machine to 31 seconds.