On 03/29/2016 06:39 PM, Carlos O'Donell wrote:
On 03/29/2016 08:22 AM, Florian Weimer wrote:
> On 03/29/2016 01:20 PM, notifications(a)fedoraproject.org wrote:
>> error building package (arch i686), mock exited with status 1; see build.log for
more information
>
> I'm trying to track this down. localedef appears quite broken. I don't
> know yet if it's an upstream change in glibc, or something in GCC.
I saw this also.
Per our discussion on IRC this looks like a vfprtinf miscompilation
where the jump tables have the wrong offsets.
Given that the jump tables are in .rodata we feel this is either a
static linker or compiler issue (or glibc issue in that it uses a
feature in the wrong way).
The code in question uses a non-standard GCC extension '&&label' to
get the address of a label. So this might be a weird thing people
didn't think about and broke.
Jakub managed to bisect to the revision causing the regression using my
slightly bizarre test case (pre-processed file plus a shell script
driving GDB to check the generated jump table), and was able to achieve
a stunning reduction of the test case. We now have a GCC PR, and a fix
is being worked on:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70460
Florian