Skip to content

Conversation

github-actions[bot]
Copy link
Contributor

@github-actions github-actions bot commented Oct 14, 2025

Backport of #120459 to release/10.0

/cc @davidwrighton

Customer Impact

  • Customer reported
  • Found internally

[Select one or both of the boxes. Describe how this issue impacts customers, citing the expected and actual behaviors and scope of the issue. If customer-reported, provide the issue number.]

Regression

  • Yes
  • No

There is a regression between .NET 9 and .NET 10 where the number of types loaded on startup has increased significantly due to greater use of static virtual methods and generics. This PR addresses the issue by changing the behavior of the type loader to not require as many instantiations of interface types to be loaded by using a variety of optimization techniques. This is not a regression in the correctness of the type loader, or in its behavior, but the overall issue being addressed is a regression.

Notably

  1. Using a technique where we load placeholder types instead of the exact needed instantiations into the interface map
  2. Changing the constraint loading code to avoid loading the constraint instantiated over the formal generic parameters of the type/method which defines the constraint
  3. In the default interface method and virtual static method scenarios reducing the number of types loaded while attempting to find the actual implementing method.

Testing

The positive effect of this change was validated by a new series of performance tests written specifically to address this issue. Testing has shown that this change reduces startup time of small applications by as much as 20%. The correctness of this change was validated by running the various tests available to run in CI. The issue here was missed previously as the startup numbers under our pre-existing startup testing scenario did not restrict CPU, and we had also been making other changes to the runtime which resulted in small applications have a comparable startup time to older versions when not run in a mode where the CPU time available to the application was restricted.

Risk

High. This change affects all types loaded into the process and adds new logic and failure points to all interface casting. In addition, this change adds additional failure modes into customer applications such that a TypeLoadException may be thrown at unexpected moments due to invalid metadata instead of the exception occurring at the current locations.

IMPORTANT: If this backport is for a servicing release, please verify that:

  • The PR target branch is release/X.0-staging, not release/X.0.

Package authoring no longer needed in .NET 9

IMPORTANT: Starting with .NET 9, you no longer need to edit a NuGet package's csproj to enable building and bump the version.
Keep in mind that we still need package authoring in .NET 8 and older versions.

…erly - Reduce the set of constraints that need to be loaded for Bounds and cast checking

TODO:
There is a path in InitTypeContext which I have #ifdef'd out as I don't understand the comment
The accessibility checking needs full loading now, as we don't have a scheme to just look at all the typedefs in a signature, and instead need to do a load of the full type and work from there. This could be revisited.
This is DRAFT as we now have an unused state WhichConstraintsToLoad::TypeOrMethodVarsOnly which we never use. (It could be used for the Bounds algorithm, but that isn't actually used without doing the more expensive logic which does access validation)
…rrently it only handles interfaces defined on valuetypes - Change it to work for interfaces that are required implementation of other interfaces - (In that case, if the interface is generic, the special instantiation type is the first type parameter of the interface, not anything else.)

This change removes some safety checks that I can't find the rationale for. This may cause some entertaining failures in CI
… - Tweak the logic so that it will attempt to avoid loading types if they are cannot have implementations of the method we're looking for
…This test was no longer actually forcing the type which is supposed to throw a TypeLoadException to load within the test - Tweak the test to actually force the offending type to be loaded
… parameters to the TypeValidationChecker in crossgen2 - Add support for doing variance safety checks to the type loader in crossgen2 - Remove the circularity of type variables checks happening in generic method load, as it is redundant - Put all of the variance safety checks in the runtime under the SkipTypeValidation flag as crossgen2 can now do it reliably
… - Add test for constraints with function pointers in them
@davidwrighton davidwrighton added the NO-MERGE The PR is not ready for merge yet (see discussion for detailed reasons) label Oct 14, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

NO-MERGE The PR is not ready for merge yet (see discussion for detailed reasons)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant