mbox series

[0/2] Make PowerPC cross tools hash the same across tunes

Message ID 20261002211037.745952-1-luciano.dittgen@ossystems.com.br
Headers show
Series Make PowerPC cross tools hash the same across tunes | expand

Message

Luciano Dittgen Oct. 2, 2026, 9:10 p.m. UTC 
yocto-check-layer test_machine_signatures fails on meta-freescale when
PowerPC machines on different tunes share a cross tool: the host-side
*-cross-powerpc recipes get different signatures although they build the
same thing. In both cases the value is identical and only its expression
differs, so the fix is the vardepvalue idiom 39bfa0dd32 already applied
to gcc-, binutils- and gdb-cross for this same check.

Tested with `bitbake -S lockedsigs` on gcc/binutils/gdb/clang-cross for
p1020rdb (e500v2), p2041rdb (e500mc), t1024rdb (e5500), t4240rdb
(e6500), p5040ds-64b and t2080rdb-64b, on master: 74 conflicting tasks
before, 29 after. All 29 left are multilib-related:

- t4240rdb sets MULTILIBS, so RECIPE_SYSROOT's expression gains
  ${MLPREFIX}; meta-freescale's to fix, not core's.
- lib32-gcc-cross-powerpc still hashes lib32-linux-libc-headers, which
  is tuned per CPU: the gcc-cross-${TARGET_ARCH}->linux-libc-headers
  entry in SIGGEN_EXCLUDE_SAFE_RECIPE_DEPS does not match the multilib
  pair, and TARGET_ARCH differs between the base and the lib32 variant.
  I don't see a clean way to express that exclusion; suggestions welcome.

Luciano Dittgen (2):
  clang-cross: Ignore how TARGET_ARCH is computed
  arch-powerpc: Hash ABIEXTENSION by its value

 meta/conf/machine/include/powerpc/arch-powerpc.inc | 3 +++
 meta/recipes-devtools/clang/clang-cross_git.bb     | 3 +++
 2 files changed, 6 insertions(+)