Allow devirtualizing more interface calls #82927
Conversation
There is an existing implementation of "does this introduce variance complexities"?
ghost
assigned
|
Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas Issue DetailsThere is an existing implementation of "does this introduce variance complexities"? Cc @dotnet/ilc-contrib
|
|
/azp run runtime-extra-platforms |
|
Azure Pipelines successfully started running 1 pipeline(s). |
| // no real upper bound on the number of actual classes implementing it | ||
| // due to MakeGenericType. | ||
| if (CanAssumeWholeProgramViewOnInterfaceUse(baseInterface)) | ||
| if (CanAssumeWholeProgramViewOnInterfaceUse(factory, type, baseInterface)) |
There was a problem hiding this comment.
Is this condition just a performance optimization that is trying to keep _disqualifiedInterfaces set small? In other words, if we deleted this condition, would anything break?
There was a problem hiding this comment.
Yes, I stared at it a bit and it doesn't look necessary. It wasn't a conscious perf optimization because it doesn't seem worth it. Deleted.
| // Interfaces implemented by arrays and array enumerators have weird casting rules | ||
| // due to array covariance (string[] castable to object[], or int[] castable to uint[]). | ||
| // Disqualify such interfaces. | ||
| if (!type.IsCanonicalSubtype(CanonicalFormKind.Any) && |
There was a problem hiding this comment.
Is the IsCanonicalSubtype early out correct? Consider:
var a = new int[1];
var b = (X<string>.MyEnum[])(object)a;
Console.WriteLine(b);
class X<T>
{
public enum MyEnum
{
}
}
There was a problem hiding this comment.
It was fine for that - we were disqualifying template types above (IsCanonical detects template types).
I moved these two checks together. It looks better.
|
/azp run runtime-extra-platforms |
|
Azure Pipelines successfully started running 1 pipeline(s). |
2 participants
There is an existing implementation of "does this introduce variance complexities"?
Cc @dotnet/ilc-contrib