This updates generated binaries and corresponding test cases up to date
after applying FixFunctionBitcasts pass.
Details
Diff Detail
- Repository
- rL LLVM
Event Timeline
This does not update these 'invalid' test cases:
test/tools/llvm-objdump/Inputs/corrupt-section.wasm test/Object/Inputs/WASM/missing-version.wasm test/Object/Inputs/WASM/string-outside-section.wasm
Because it is unclear how to regenerate these invalid test cases. Do you think we should regenerate them? If so, how? @sbc100
Hmm.. just to be clear, these tests are not failing on HEAD are they?
Why not just leave them as is?
Also, I've had some fallout from the change to enable the main wrappers by default so I think we should perhaps revert it. Specifically the test_mainenv test in emscripten is failing because it expects envp to be set to null, not some undefined value (which happens to be non-null right now).
Before test/Object/Inputs/WASM/string-outside-section.wasm failed after I added the event section, but maybe it is because I reassigned section number incorrectly.
Why not just leave them as is?
Yes I'm gonna leave them as is for now.
Also, I've had some fallout from the change to enable the main wrappers by default so I think we should perhaps revert it. Specifically the test_mainenv test in emscripten is failing because it expects envp to be set to null, not some undefined value (which happens to be non-null right now).
I rebased this CL on top of D54117. We apparently still have something to update because of FixFunctionBitCasts on SomeOtherFunction.
Its not clear to me that we should regenerate these binaries every time a change to llvm means the binary would change. These test inputs remain valid even if llvm would no longer that same file bit-for-bit.
I wasn't planning on regenerating test binaries every time; the reason I did this patch was, after I incorrectly changed section numbers (by making the event section 7 and shifting all later sections by 1), I had to regenerate test binaries, and accidentally all these FixFunctionBitCast changes were included in that CL. Anyway, after I reassigned the correct section number (12) to the event section, I don't need to update binary tests anymore there so I don't need this CL either. Thanks!
Hmm, come to think about it, maybe it's better to update the binaries here and in D54096 after all, because D54096 adds a new section. (I'm not planning on updating these every time we change something on binaries) The reason is, for example, if I want to see if the event section code (0xC) is correctly printed in this test, I have to update binaries. And if I update binaries only in D54096 and abandon this CL, these changes from FixFunctionBitCasts will be all included in that CL, which I don't want.