Re: x86_64-solaris

[Rainer Orth]

Earnie <address@hidden> writes:

> On 12/19/2018 5:33 AM, Rainer Orth wrote:
>> Hi Karl,
>>
>> I've just investigated a GCC bug report about a comparison failure on
>> Solaris/SPARC:
>>
>>      PR target/88535
>>          sparcv9 gcc 7 causes comparison failure in sparc gcc 8 dwarf2out.o
>>          https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88535
>>
>> where config.guess failed to notice that the build compiler was
>> 64-bit-default, while the host/target was 32-bit-default.
>>
>> This doesn't happen with the config.guess/config.sub currently bundled
>> with gcc mainline (2018-06-26 resp. 2018-07-03) where with a
>> 64-bit-default gcc x86_64-pc-solaris2.11 is determined, while with
>> CC='gcc -m32' I get i386-pc-solaris2.11 just as expected.
>>
>
> The proper use of configure can help eliminate the need for config.guess
> and config.sub to interfere here.  Use --build and --host as appropriate;
> and at times you'll need --target depending on what you're building.

While true, that's very easy to get wrong, especially in multilib cases
like Solaris.  The config.guess change I'm complaining about is almost
guaranteed to break all sorts of toolchain builds (gcc, binutils, and
gdb), as the GCC bug report above clearly shows, and it took several
people quite some time to diagnose what was wrong.  Configurations that
worked out of the box before on Solaris/x86 now become broken once
current config.guess/config.sub get imported into gcc.

That's both a waste of developer time and a terrible user experience.

        Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University