Skip to content

Conversation

@MichalStrehovsky
Copy link
Member

@MichalStrehovsky MichalStrehovsky commented Jul 8, 2025

Fixes #103951.

Same method with different genericness are not distinguishable in the stack traces. They are still distinguishable in the debugger, but it might be acceptable, especially since we have an opt out (undocumented for now).

Cc @dotnet/ilc-contrib

Fixes dotnet#103951.

Same method with different genericness are not distinguishable in the stack traces. They are still distinguishable in the debugger, but it might be acceptable, especially since we have an opt out (undocumented for now).
@dotnet-policy-service
Copy link
Contributor

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

@jkotas
Copy link
Member

jkotas commented Jul 8, 2025

They are still distinguishable in the debugger, but it might be acceptable, especially since we have an opt out (undocumented for now).

Do you think it would be possible to rename the symbol for folded copy such that the symbol name does not include the instantiation that happened to win?

@MichalStrehovsky
Copy link
Member Author

They are still distinguishable in the debugger, but it might be acceptable, especially since we have an opt out (undocumented for now).

Do you think it would be possible to rename the symbol for folded copy such that the symbol name does not include the instantiation that happened to win?

I think it would make things quite complicated in general. We don't have facilities to change our mind about how symbols are named. We'd also need to invent a scheme for new stable names.

@MichalStrehovsky MichalStrehovsky marked this pull request as ready for review July 8, 2025 08:35
@Copilot Copilot AI review requested due to automatic review settings July 8, 2025 08:35
@MichalStrehovsky
Copy link
Member Author

Size statistics

Pull request #117411

Project Size before Size after Difference
TodosApi-linux 25538640 24666192 -872448
TodosApi-windows 22906368 22239232 -667136
avalonia.app-linux 19995056 19499440 -495616
avalonia.app-windows 17649664 17309696 -339968
hello-linux 1369040 1364944 -4096
hello-minimal-linux 1086248 1086248 0
hello-minimal-windows 837120 837120 0
hello-windows 1092608 1088512 -4096
kestrel-minimal-linux 5205784 5205784 0
kestrel-minimal-windows 4750848 4750848 0
reflection-linux 1992752 1964080 -28672
reflection-windows 1695232 1673728 -21504
webapiaot-linux 9866600 9653608 -212992
webapiaot-windows 9032704 8862720 -169984
winrt-component-full-windows 5720064 5594112 -125952
winrt-component-minimal-windows 1701888 1701888 0

Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull Request Overview

This PR replaces the boolean “fold identical method bodies” flag with a three-way MethodBodyFoldingMode enum (None, Generic, All), defaults generic folding by build integration, and wires the new mode through the compiler and MSBuild targets.

  • Swapped UseMethodBodyFolding(bool) for UseMethodBodyFolding(MethodBodyFoldingMode) in the CLI, builder, and RyuJit compilation paths.
  • Introduced MethodBodyFoldingMode enum and revamped ObjectDataInterner to only fold based on generics when requested.
  • Updated MSBuild targets to compute a default folding mode and emit the new --methodbodyfolding:<mode> argument.

Reviewed Changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file
File Description
src/coreclr/tools/aot/ILCompiler/Program.cs Parse string option into new MethodBodyFoldingMode
src/coreclr/tools/aot/ILCompiler/ILCompilerRootCommand.cs Change CLI option type from bool to string and update description
src/coreclr/tools/aot/ILCompiler.RyuJit/Compiler/RyuJitCompilationBuilder.cs Switch interner creation to use new enum
src/coreclr/tools/aot/ILCompiler.Compiler/Compiler/ObjectDataInterner.cs Add _genericsOnly flag, CanFold, and comparer support
src/coreclr/tools/aot/ILCompiler.Compiler/Compiler/DependencyAnalysis/NodeFactory.cs Use CanFold instead of IsNull
src/coreclr/tools/aot/ILCompiler.Compiler/Compiler/CompilationBuilder.Aot.cs Change builder API and add enum definition
src/coreclr/nativeaot/BuildIntegration/Microsoft.NETCore.Native.targets Compute and pass <mode> to --methodbodyfolding
Comments suppressed due to low confidence (3)

src/coreclr/tools/aot/ILCompiler/ILCompilerRootCommand.cs:103

  • The description is missing a closing parenthesis; change to e.g. "Fold identical method bodies (one of: none, generic, all)".
            new("--methodbodyfolding") { Description = "Fold identical method bodies (one of: none, generic, all" };

src/coreclr/tools/aot/ILCompiler/ILCompilerRootCommand.cs:102

  • Add an AddValidator to restrict the option to the valid values (none, generic, all) and produce a clear error when parsing fails.
        public Option<string> MethodBodyFolding { get; } =

src/coreclr/tools/aot/ILCompiler/ILCompilerRootCommand.cs:104

  • No tests appear to validate the new MethodBodyFolding modes; please add unit tests for parsing and default behavior to ensure coverage.
        public Option<string[]> InitAssemblies { get; } =

@MichalStrehovsky
Copy link
Member Author

@sbomer is this something you'd be comfortable reviewing?

Copy link
Member

@sbomer sbomer left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Did you consider making the defaults dependent on DebuggerSupport?

@MichalStrehovsky
Copy link
Member Author

Did you consider making the defaults dependent on DebuggerSupport?

I didn't think about it but gave it a thought now - here's my thinking:

  • DebuggerSupport disables debugger support. For example, it controls whether we leave the breadcrumbs for the SOS extension to work. The SOS extension doesn't work at all with DebuggerSupport=false. Debugging is much harder.
  • Method folding somewhat messes up the callstack, but not to the extent that it would be very difficult to reason about. It's more to the tune of how inlining and tail calls mess up callstack.

So I'd lean towards not tying it to that.

@MichalStrehovsky MichalStrehovsky merged commit 0621e64 into dotnet:main Jul 18, 2025
95 checks passed
@MichalStrehovsky MichalStrehovsky deleted the foldgeneric branch July 18, 2025 08:21
@github-actions github-actions bot locked and limited conversation to collaborators Aug 17, 2025
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Investigate folding identical methods with genericness by default

4 participants