Add an interesting test case for compiler generated code marking #2928
Conversation
See the test for description of what it does. It's basically the linker repro for the case hit in dotnet/runtime#73027 (comment)
There was a problem hiding this comment.
Interesting, thanks!
Is the warning behavior what you would expect? In theory the descriptor marking should produce the same kinds of warnings as reflection access to compiler-generated code, but that might be too noisy.
So the fix would be to make the processing of both cases consistent, but it wouldn't change the warnings or the output, right?
|
Right - the behavior currently seems OK to me - we don't produce warning in either case. Basically even though the entire assembly is preserved, we treat the compiler generated code as "trimmable" anyway (we would still keep it, but we won't produce diagnostics from it). I think that's OK, less noise and I don't see any holes (we don't see any reference to it anywhere). The internal behavior is obviously weird - it would be great if we could make it to be consistent regardless of ordering. But I could not find any observable differences yet. |
…net/linker#2928) See the test for description of what it does. It's basically the linker repro for the case hit in #73027 (comment) Commit migrated from dotnet/linker@463997e
See the test for description of what it does. It's basically the linker repro for the case hit in dotnet/runtime#73027 (comment)