[OpenAFS-devel] buildbot maintenance

[Simon Wilkinson]

On 3 Sep 2012, at 16:49, Jason Edgecombe wrote:
> The gcc upgrade (4.4.5) seems to have uncovered some compilation =
issues:
> make[3]: Entering directory `/home/buildmaster/git/src/tptserver'
> gcc -O2 -Wall -Wstrict-prototypes -Wold-style-definition -Werror =
-fdiagnostics-show-option -Wpointer-arith =
-I/home/buildmaster/git/src/config -I/home/buildmaster/git/include -I. =
-I. -pthread -D_REENTRANT -DAFS_PTHREAD_ENV -o ptutils.o -c =
./../ptserver/ptutils.c
> cc1: warnings being treated as errors
> ./../ptserver/ptutils.c: In function =91ChangeEntry=92:
> ./../ptserver/ptutils.c:1964: error: dereferencing pointer =91tent.243=92=
 does break strict-aliasing rules [-Wstrict-aliasing]
> ./../ptserver/ptutils.c:1964: note: initialized from here
> ./../ptserver/ptutils.c: In function =91RemoveFromSGEntry=92:
> ./../ptserver/ptutils.c:813: error: dereferencing pointer =91tentryg=92 =
does break strict-aliasing rules [-Wstrict-aliasing]
> ./../ptserver/ptutils.c:798: error: dereferencing pointer =91tentryg=92 =
does break strict-aliasing rules [-Wstrict-aliasing]
> ./../ptserver/ptutils.c:767: error: dereferencing pointer =91tentryg=92 =
does break strict-aliasing rules [-Wstrict-aliasing]
> ./../ptserver/ptutils.c:766: error: dereferencing pointer =91tentryg=92 =
does break strict-aliasing rules [-Wstrict-aliasing]
> ./../ptserver/ptutils.c:763: note: initialized from here
> make[3]: *** [ptutils.o] Error 1
> make[3]: Leaving directory `/home/buildmaster/git/src/tptserver'
> make[2]: *** [tptserver] Error 2
> make[2]: Leaving directory `/home/buildmaster/git'
> make[1]: *** [build] Error 2
> make[1]: Leaving directory `/home/buildmaster/git'
> make: *** [all] Error 2

This looks like the same problems with aliasing when supergroups is =
enabled that f2db78a346112f5320efc6f0b6b9fe4ae0d893d3 fixed in the =
non-pthreaded build. I've tried pushing a fix to gerrit (8035) which =
copies the same build rules that we use to work around this in the =
ptserver directory to the tptserver one, and we'll see if that hides the =
problem.

At some point, we're going to have to bite the bullet and rework the =
supergroups code so that we don't have these issues, but that's going to =
require significant changes, as the supergroups code plays fast and =
loose with structure definitions.

Cheers,

Simon.