@am11 Since I think you worked in this area the most, what do you think about this? It seems that the GCC build job failed cause ld does not support the flag.

Do the official builds use lld instead? If not, it probably doesn't make sense to do any checking for linker here.

Also, it seems like NativeAOT doesn't use the flags from here.

@MichalStrehovsky, experimented with it earlier. We may need changes from #87045.

@MichalStrehovsky, experimented with it earlier. We may need changes from #87045.

My intent here was to only enable it for native code in the runtime, I'd leave the handling of NAOT outputs to @MichalStrehovsky.

Do you see any size difference before/after on linux?

$ ./build.sh clr+libs -c Release
$ find artifacts/bin/coreclr/linux.x64.Release -name '*.so*' -exec du -sh {} \;

Do you see any size difference before/after on linux?

$ ./build.sh clr+libs -c Release
$ find artifacts/bin/coreclr/linux.x64.Release -name '*.so*' -exec du -sh {} \;

I currently do not have a Linux machine on hand to test it, that's why I wanted to see how it impacts rt-sz.

@am11 Can you show the total without .dbg files?

Also, it seems like emscripten doesn't support icf even with lld, what's the proper condition for excluding that?

@am11 Can you show the total without .dbg files?

| 157295704 | 154012152 | -3283552 | -2.09% | TOTAL |

captured befroe.md and after.md with this:

echo "| Size (human) | Size (bytes) | Path |"; find artifacts/bin/coreclr/linux.arm64.Release artifacts/bin/native/net11.0-linux-Release-arm64 -name '*.so*' | xargs -I{} bash -c 'echo "| $(du -sh "$0" | cut -f1) | $(du -sb "$0" | cut -f1) | $0 |"' {}

then asked AI to compute deltas (which used python script).

Tagging subscribers to this area: @dotnet/runtime-infrastructure
See info in area-owners.md if you want to be subscribed.

/azp run runtime-coreclr outerloop, runtime-extra-platforms

Failures all unrelated, either non Linux or networking errors.

@jkotas Do you know who'd be the right person to review this?

This change is not risk-free. --icf=all violates the C/C++ spec guarantees about pointer equality. E.g. from the C23 spec: "Two pointers compare equal if and only if both are null pointers, both are pointers to the same object
(including a pointer to an object and a subobject at its beginning) or function, ...".

We have the equivalent option enabled on Windows, and copilot is not able to find any dependencies on the pointer equality behavior in non-Windows code, so the risk should be acceptable.

Thanks!