[PATCH v2 0/6] Fixes for the parallel make/make check support

[Sam James]

Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:

> This series fixes fallouts from the parallel subdirectory recursion
> and parallel 'make check' changes that went into 2.44 (7cac99621e96
> and related commits).
>
> The first patch fixes gen-as-const-headers races where a generated
> header is consumed by a different subdirectory than the one that
> generates it (BZ 34438).  The serial recursion hid the missing
> dependency because the sorted subdirectory order always ran the
> generating subdirectory first; the parallel fan-out does not provide
> that ordering.  The headers are moved to the subdirectories that
> consume them.
>
> The second patch fixes a parallel 'make install' failure where
> install-others-nosubdir adds the ABI lib-names header as a
> prerequisite in subdirectories without $(headers), where no install
> rule for it exists (BZ 34439).  This also only worked because
> .NOTPARALLEL previously guaranteed the top level installed the header
> first.
>
> The third patch fixes the two-pass 'make check' so the static checks
> (check-abi, conformtest, installed headers) run in the first
> build-only pass no longer abort the check before any built test runs,
> and the test summary is only merged and printed once at the end of
> the second pass, restoring the single complete summary.
>
> The fourth patch makes the .dt to .d dependency conversion safe
> against concurrent sub-makes.
>
> The last two patches fix the remaining races observed by Ross Burton.
> The fifth creates the top-level stamp files before the subdirectory
> fan-out, and the the sixth orders the arm rtld link after
> libgcc-stubs.a.
>
> Tested with builds and 'make check' on x86_64-linux-gnu, and builds
> for sh4-linux-gnu, i686-gnu, x86_64-gnu, and arm-linux-gnueabihf.
>
> I also run multiple builds on cfarm120, which has 192 cores and
> most likely to trigger concurrency issues (as Florian has also
> noticed).

Thanks for respinning with the extra fixes.

>
> Adhemerval Zanella (5):
>   Fix gen-as-const-headers races with the parallel subdir recursion (BZ
>     34438)
>   Makefile: Only print the test summary in the second pass of 'make
>     check'
>   Makerules: Make the .dt to .d conversion safe against concurrent
>     sub-makes
>   Makefile: Order the top-level stamp files before the subdirectory
>     fan-out
>   arm: Order the rtld link after libgcc-stubs.a
>
> Rudi Heitbaum (1):
>   Makerules: Only install the ABI lib-names header from the top level
>     (BZ 34439)
>
>  Makefile                                | 15 ++++++++
>  Makerules                               | 46 +++++++++++++++++++++----
>  sysdeps/arm/Makefile                    |  5 +++
>  sysdeps/mach/hurd/x86/Makefile          |  6 +---
>  sysdeps/unix/sysv/linux/sh/Makefile     |  4 ++-
>  sysdeps/unix/sysv/linux/x86_64/Makefile |  4 ++-
>  sysdeps/x86/Makefile                    | 10 ++++--
>  7 files changed, 74 insertions(+), 16 deletions(-)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 418 bytes
Desc: not available
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20260806/199d3c76/attachment.sig>