[SYCL] Fix compilation issue introduced by d0d8a56

[Naghasan]

fix compilation issue introduced by #15061

[Naghasan]

@sarnex the fix

[sarnex]

thx will merge once build passes

[sarnex]

build passed, merging early because change is trivial and build is broken on multiple platforms. will monitor postcommit.

[Naghasan]

the error is caused by Werror, shouldn't that be used as well in the precommit build ? I understand why there is post commit tests, but I feel my failure could have been caught before.

[sarnex]

it is used in precommit, we build both with -DSYCL_ENABLE_WERROR=ON

the problem is the compiler is different

precommit

2024-12-04T11:58:58.5567290Z   cc: gcc
2024-12-04T11:58:58.5567773Z   cxx: g++
2024-12-04T11:58:58.5568315Z   build_image: ghcr.io/intel/llvm/ubuntu2204_build:latest
2024-12-04T11:58:58.5568921Z   build_ref: 4b3842901ccba38a6a268fdafb4bfaf2dbf625fe

postcommit

2024-12-04T14:25:56.9074558Z   cc: clang
2024-12-04T14:25:56.9074850Z   cxx: clang++
2024-12-04T14:25:56.9075230Z   build_image: ghcr.io/intel/llvm/sycl_ubuntu2204_nightly:build

so it seems like gcc's werror is much more lenient. maybe we should also use clang in precommit, and add a nightly job to build in gcc just to make sure it works? opinions @aelovikov-intel @uditagarwal97

[sarnex]

@Naghasan Build is still broken with seemingly the same error

2024-12-04T16:30:59.4365706Z /__w/llvm/llvm/src/sycl/source/detail/scheduler/commands.cpp:2512:56: error: suggest braces around initialization of subobject [-Werror,-Wmissing-braces]
2024-12-04T16:30:59.4366308Z  2512 |         {UR_EXP_LAUNCH_PROPERTY_ID_WORK_GROUP_MEMORY, {WorkGroupMemorySize}});

https://github.com/intel/llvm/actions/runs/12164078580/job/33924842681

Can you build with clang when verifying your fix locally?

[aelovikov-intel]

maybe we should also use clang in precommit, and add a nightly job to build in gcc just to make sure it works? opinions @aelovikov-intel @uditagarwal97

I think the cases like this one are rare enough and the current split between pre/post does the right job. Also, why would anyone use gcc for their normal day-to-day work anyway? ;)

[sarnex]

should we switch it so we use clang in precommit and gcc in post?

[uditagarwal97]

should we switch it so we use clang in precommit and gcc in post?

If clang's werror is stricter, it makes sense to use clang in pre-commit and gcc in post.

[sarnex]

ive never seen a werror issue in precommit, so that seems to be the case. ill try switching it.

[aelovikov-intel]

I don't think is strictly "stricter" than the other. Wouldn't be surprised if we'll have same kind of situation after the switch.

Anyhow, up to @sarnex and @uditagarwal97 to decide as they're the ones keeping our CI running (and many thanks to them for that!)

[sarnex]

thanks, ill try it and monitor it and if we see more postcommit fails the other way ill switch it back

[aelovikov-intel]

While on it, do you think we can add a windows build with OneAPI release (vs MSVC) in post or nightly?

[sarnex]

yeah let me try (gonna regret this prolly)

[sarnex]

im fixing the build in #16268

[Naghasan]

thanks @sarnex , apologies I was trapped in a meeting

[sarnex]

np

[sarnex]

These kind of issues should be caught in precommit now because of 5e0db3e