[libc] Fix GPU test build error

[jplehr]

This fixes a build error on the AMDGPU buildbot introduced in PR #92172

[llvmbot]

@llvm/pr-subscribers-libc

Author: Jan Patrick Lehr (jplehr)

Changes

This fixes a build error on the AMDGPU buildbot introduced in PR #92172

---

Full diff: https://github.com/llvm/llvm-project/pull/92235.diff

1 Files Affected:

• (modified) libc/src/__support/StringUtil/tables/stdc_errors.h (+1-2)

diff --git a/libc/src/__support/StringUtil/tables/stdc_errors.h b/libc/src/__support/StringUtil/tables/stdc_errors.h
index a9c1527834550..6873d6bd51074 100644
--- a/libc/src/__support/StringUtil/tables/stdc_errors.h
+++ b/libc/src/__support/StringUtil/tables/stdc_errors.h
@@ -15,11 +15,10 @@
 
 namespace LIBC_NAMESPACE {
 
-LIBC_INLINE_VAR constexpr const MsgTable<4> STDC_ERRORS = {
+LIBC_INLINE_VAR constexpr const MsgTable<3> STDC_ERRORS = {
     MsgMapping(0, "Success"),
     MsgMapping(EDOM, "Numerical argument out of domain"),
     MsgMapping(ERANGE, "Numerical result out of range"),
-    MsgMapping(EILSEQ, "Invalid or incomplete multibyte or wide character"),
 };
 
 } // namespace LIBC_NAMESPACE

[lntue]

Look like you also need to update libc/test/src/string/strerror_test.cpp:100:

[3952/4270] Running unit test libc.test.src.string.strerror_test.__unit__
FAILED: projects/libc/test/src/string/CMakeFiles/libc.test.src.string.strerror_test.__unit__ /var/lib/buildkite-agent/builds/linux-56-59b8f5d88-2sq4t-1/llvm-project/github-pull-requests/build/projects/libc/test/src/string/CMakeFiles/libc.test.src.string.strerror_test.__unit__
cd /var/lib/buildkite-agent/builds/linux-56-59b8f5d88-2sq4t-1/llvm-project/github-pull-requests/build/projects/libc/test/src/string && /var/lib/buildkite-agent/builds/linux-56-59b8f5d88-2sq4t-1/llvm-project/github-pull-requests/build/projects/libc/test/src/string/libc.test.src.string.strerror_test.__unit__.__build__
[ RUN      ] LlvmLibcStrErrorTest.KnownErrors
/var/lib/buildkite-agent/builds/linux-56-59b8f5d88-2sq4t-1/llvm-project/github-pull-requests/libc/test/src/string/strerror_test.cpp:154: FAILURE
      Expected: __llvm_libc_19_0_0_git::strerror(static_cast<int>(i))
      Which is: Unknown error 84
To be equal to: message_array[i]
      Which is: Invalid or incomplete multibyte or wide character
[  FAILED  ] LlvmLibcStrErrorTest.KnownErrors

[michaelrj-google]

We will need to figure out a more permanent solution for errno numbers on GPUs because I would like to revert this patch when possible. Having a valid EILSEQ strerror on platforms that support it is important.

[jhuber6]

We will need to figure out a more permanent solution for errno numbers on GPUs because I would like to revert this patch when possible. Having a valid EILSEQ strerror on platforms that support it is important.

I think this was broken on the CPU build beforehand, at least it was when I checked it, so reverting this would make that fail again.

[robincaloudis]

We will need to figure out a more permanent solution for errno numbers on GPUs because I would like to revert this patch when possible. Having a valid EILSEQ strerror on platforms that support it is important.

I think this was broken on the CPU build beforehand, at least it was when I checked it, so reverting this would make that fail again.

This is true. #92172 broke fuchsia-x86_64-linux and openmp-offload-libc-amdgpu-runtime. I have not yet fully understood all details of the problem. I keep you updated.