Re: [gentoo-soc] [[e-mail]: A question about BUILD_DIR variable in cmake.eclass under EAPI=8]

[Michał Górny]

On Sat, 2022-07-16 at 17:38 +0800, wuyy wrote:
> Hello all,
>
> Below is the my question about BUILD_DIR variable in cmake.eclass. I
> have previously raised it in IRC channel, but did not get it clear, so I
> tried to ask the author of cmake.eclass. It seems that Andreas may be
> busy recently and I haven't receive a reply, and I'll keep wait. Thus, I
> forward this question here so we can discuss it and keep a public
> record. If the problem is confirmed to be rather important, I think we
> can move the discussion to Gentoo-dev mailing list.
>
> ----- Forwarded message from wuyy <xgreenlandforwyy@×××××.com> -----
>
> Date: Fri, 15 Jul 2022 16:49:45 +0800
> From: wuyy <xgreenlandforwyy@×××××.com>
> To: Andreas Sturmlechner <asturm@g.o>
> Subject: A question about BUILD_DIR variable in cmake.eclass under EAPI=8
>
> Hello Andreas,
>
> I'm a student participating in Google Summer of Code this year,
> packaging ROCm. I have met a problem when using cmake.eclass, and I'm
> not sure whether it should be called a bug or it's just a feature, so I
> want to personally ask you, the author of corresponding code.
>
> As I observe the BUILD_DIR variable is no longer globally set in
> cmake.eclass under EAPI=8, with this change:
>
> -# ${WORKDIR}/${P}_build.
> -: ${BUILD_DIR:=${WORKDIR}/${P}_build}
> +# ${CMAKE_USE_DIR}_build (in EAPI-7: ${WORKDIR}/${P}_build).
> +[[ ${EAPI} == 7 ]] && : ${BUILD_DIR:=${WORKDIR}/${P}_build}
> +# EAPI-8: set inside _cmake_check_build_dir
>
> So I have to use this variable in ebuild after _cmake_check_build_dir
> which is inside cmake_src_prepare. That's not an issue in src_prepare.
> Also this variable can be used in following phases.  Everything seems
> nice, until I begin debugging and polishing src_test.  After a
> successful src_compile, I come to the src_test, which has bugs.  I fixed
> the bug and rerun ebuild xxx src_test, but it unexpectedly failed,
> because BUILD_DIR is empty!
>
> To summarize, BUILD_DIR can live to other phases when phases are
> executed in one call, but cannot be kept if the previous steps are
> skipped.
>
> This is not a problem for most scenarios, but it's a bit annoying when
> developing and debugging ebuilds. I have to call _cmake_check_build_dir
> explicitly before using BUILD_DIR, although src_prepare has been run. I
> wonder this may be a bug, but maybe it's designed to work that way for
> some reasons I don't know. If you think this is a bug, I can put it on
> bugzilla and find a fix to it. Thank you very much!
>

Can you reproduce while running the test phase via emerge(1)?  Using
ebuild(1) is known to be buggy.

-- 
Best regards,
Michał Górny