Size regression in Minimal API NativeAOT app #82732
Description
Description
Between builds 8.0.0-preview.2.23123.4+ff7c19f and 8.0.0-preview.2.23123.5+5ba867f there was a ~140KB size regression on linux-x64 Minimal API NativeAOT app.
Configuration
- Which version of .NET is the code running on? .NET 8
- What OS version, and what distro if applicable? Linux
- What is the architecture (x64, x86, ARM, ARM64)? x64
Regression?
Yes. From a previous build.
Data
Here are the crank commands to repro:
crank --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/scenarios/goldilocks.benchmarks.yml --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/build/ci.profile.yml --scenario basicminimalapipublishaot --profile intel-load2-app --profile amd-lin2-load --profile amd-lin2-db --application.packageReferences Microsoft.Dotnet.ILCompiler=8.0.0-preview.2.23123.4 --application.framework net8.0 --application.options.collectCounters true --application.aspNetCoreVersion 8.0.0-preview.2.23123.5 --application.runtimeVersion 8.0.0-preview.2.23123.4 --application.sdkVersion 8.0.100-preview.2.23124.1
and
crank --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/scenarios/goldilocks.benchmarks.yml --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/build/ci.profile.yml --scenario basicminimalapipublishaot --profile intel-load2-app --profile amd-lin2-load --profile amd-lin2-db --application.packageReferences Microsoft.Dotnet.ILCompiler=8.0.0-preview.2.23123.5 --application.framework net8.0 --application.options.collectCounters true --application.aspNetCoreVersion 8.0.0-preview.2.23123.5 --application.runtimeVersion 8.0.0-preview.2.23123.5 --application.sdkVersion 8.0.100-preview.2.23124.1
Analysis
By diffing the mstat files between these two builds, it appears the majority of the size regression comes from the System namespace.
And drilling into the System namespace, I see a bunch of SZGenericArrayEnumerator`1 types that weren't there before, but are now.
And inspecting the diff between the two builds: ff7c19f...5ba867f, this leads me to believe the regression was caused by #82499. (cc @stephentoub)
I've zipped the before and after mstat files, along with a text dump in BasicMinimalApi.zip, if anyone wants to take a look.