No description
f32a45bfac
riscv64 is a moving target we are no longer chasing here, so revert every
riscv64-specific change and leave the arch on RH's upstream behaviour:
- remove gcc16-riscv64-extfloat.patch (Patch9001) entirely;
- remove the %ifarch riscv64 restore of the %build PATH to the SCL bindir;
- remove the #if !defined(__riscv) guards from gcc16-nonshared-hidden.patch,
so __to_chars_10_impl<unsigned long> falls back into RH's #else branch and
the _Sp_counted_* block in tzdb140.cc is asserted unguarded again.
What stays: the i686 work (ILP32 manglings, __get_leap_second_info,
_M_release_last_use_cold, the refreshed _M_visit manglings) and the
x86_64_v2 work (--with-arch_64=x86-64-v2 and the __AVX2__ guards). Both are
unrelated to riscv64 and both are load-bearing -- without the first the v2
toolset silently generates x86-64-v3 code, without the second the v2 build
hits the same phantom UND.
riscv64 stays in the build arch list and is expected to fail again in
configure-stage1-target-libgcc.
Patch numbering is left alone: the nonshared patch keeps 9002 even though
9001 is now unused.
Verified with gpatch -p1 --fuzz=0 against the post-compat state of
nonshared20/{format-inst,clock,tzdb140}.cc reconstructed from c10s
gcc16-libstdc++-compat.patch -- all three hunks apply with no fuzz and no
offset.
|
||
|---|---|---|
| files | ||
| config.yaml | ||