@rovka : There is one more instance of similar code in function translateInvoke around line 1670. I think you will need to update that as well.

Thanks for the review! I will check them in as two separate commits.

Confirmed that reverting the change locally fixes the test. Could you please take a look.

It seems that following test is broken with this change: llvm/unittests/IR/PassBuilderCallbacksTest.cpp

Thanks for the fix!

It seems that following test is broken with this change: DebugInfo/WebAssembly/dbg-value-live-interval.ll
Confirmed that reverting the change locally fixes the tests. Could you please take a look.

It seems that following test is broken with this change: DebugInfo/WebAssembly/dbg-value-live-interval.ll
Confirmed that reverting the change locally fixes the tests. Could you please take a look.

It seems that following tests are broken with this change:

I just saw that the change is reverted already. Thanks!

It seems that this change causes two clang tests to fail since they didn't expect this warning.

LGTM

What is the expected behavior for casting float values in range (-1.0 , 0.0) from float->int->float? If the expected value is -0.0 (instead of 0.0), what is the reason for that?

LGTM

Updated the summary for option description.

Removed the test which ensured that regular LTO file was written with index only option. It needs to have obj-path option to write that file.

Rebased diff.

This change is already in https://reviews.llvm.org/D46480. I am rebasing it.

LGTM

Rebased the diff

LGTM

Rebased diff

If we use return (instead of exit) as in this patch, the final params file( Config->ThinLTOIndexOnlyObjectsFile) is not fully generated and few of the last entries are not written(even though respective files are generated) . Here _exit() is called eventually from exitLld function call in elf::link . However, if we replace that with exit(), the file is generated properly. From here, it seems that _exit() is not able to close the file properly.

I will be sending another patch to address tejohnson's concerns. I think, in the meantime, we can add emit-imports option support. Or would you like me to send that patch first?

I will send out a patch to address both concerns. Thank you for reviewing.

I need to make similar change while creating empty index files, along with updating the description in test for it. I will add that change in next patch if that is okay.

Added createEmptyFile function to create empty import files. Also updated the test with error description.

Added test to ensure that lld doesn't generate import files when thinlto-index-only is not enabled