Re: Patch: Fix OS X PowerPC support

[Sevan Janiyan]

On 14/11/2023 18:42, Bruno Haible wrote:
> I can't find the time to think through the detailed results that you listed.
> I hope you can do that.

No worries, hopefully I wont end up being to much of a bother.

> But in summary, my point was "the patch is odd, because it uses MIN in one
> place and MAX in the other place" and I guessed "it will not work for a range
> that includes both 10.4 and 10.5". Your answer is "this will not work anyway",
> right? Did I understand it right?

Correct.

> If so, can you change your patch so that
>    1) it is symmetric w.r.t. to MIN and MAX,
> and/or
>    2) it at least compiles for a range that includes both 10.4 and 10.5,
>       and ideally works on 10.5? I think that would be the best behaviour
>       for MacPorts/Homebrew/Fink/etc.

Ok, I reworked the patch so that it only relies on 
MAC_OS_X_VERSION_MAX_ALLOWED to set SIGSEGV_FAULT_STACKPOINTER on 10.4 & 
older. It falls through on the else case to set things up on 10.5.

Tested results as follows:
builds on tiger with just ./configure, tests fail, needing libsigsegv,
test-sigsegv-catch-stackoverflow1 & 2 fail, as expected

./configure with libsigsegv on tiger builds, tests pass

./configure with CFLAGS=-mmacosx-version-min=10.5 on tiger fails
sigsegv.c: In function ‘sigsegv_handler’:
sigsegv.c:1044: error: ‘struct mcontext’ has no member named ‘__ss’

./configure with libsigsegv CFLAGS=-mmacosx-version-min=10.5 on tiger 
builds, but since it links to libsigsegv, needs that installed on 10.5 
for test binaries to run there as well.

./configure with CFLAGS=-mmacosx-version-min=10.4 on leopard builds but 
test binaries fail to run due to unresolvable symbols on tiger.



Sevan

patch-sigsegv-ppc-darwin-mk2.txt
Description: Text document