Re: Breakage on builder Fedora-x86_64-w64-mingw32, revision 49b000dc5686e016f05b717f18d2c8f865eb3617 [gdb-8.12-branch]

[Pedro Alves <palves at redhat dot com>]

On 07/13/2018 12:16 PM, Pedro Alves wrote:
> On 07/13/2018 11:55 AM, Wiederhake, Tim wrote:
>>> To: Pedro Alves <palves@redhat.com>
>>>
>>> On Thursday, July 12 2018, Pedro Alves wrote:
>>>
>>>> On 07/12/2018 08:59 AM, gdb-buildbot@sergiodj.net wrote:
>>>>> yes
>>>>> checking for memcmp... In file included from ../../binutils-
>>> gdb/intl/plural.y:35:0:
>>>>> ../../binutils-gdb/intl/plural-exp.h:102:23: error: conflicting types
>>> for 'libintl_gettextparse'
>>>>>  # define PLURAL_PARSE libintl_gettextparse
>>>>>                        ^
>>
>> For what it's worth, we're seeing this, too, every now and then.
>> Up to now I believed this was just a problem with our setup, as I couldn't find any difference in the source or build directories of failing and working builds.
>> Additionally, cross-compiling gdb seems to increase the frequency of failures, as does having the source or build directory on a network share.
> 
> My guess is that for some reason GNU Make thinks it
> needs to rebuild plural.c from src/intl/plural.y instead of
> using the in-source-tree src/intl/plural.c, and then we run into some
> incompatibility between gettext and bison (or rather,
> the newer-bison-produced file).  Network shares are prone to
> time skew, >1s differences, etc. confusing make.
> 
> Oh, hmm, I've been building my mingw cross gdbs with --disable-intl.
> If I remove that, then I see that same error too.
> 
> Yep, if I
>  touch src/intl/plural.c
> then
>  rm -f build/intl/plural.* && make -C intl
> build successfully.
> 
> I kind of have the feeling I've seen this discussed before.

Found <https://sourceware.org/ml/binutils/2014-12/msg00107.html>.

Looks like there's an intl fix, but nobody ever imported it into
our copy (starting with updating gcc's copy).

Anyone want to take a stab at it?

Thanks,
Pedro Alves