Re: unexelf on irix (kinda digest)
[ I've removed all those people from CC: who I know to be on xemacs-beta ]
Max Matveev <makc(a)sgi.com> writes:
Olivier Galibert writes:
:> On Mon, Dec 06, 1999 at 09:04:15AM -0200, Alexandre Oliva wrote:
:> > I have found out the `.got' section didn't need special handling on
:> > IRIX either, at least when the compiler was gcc. If this is confirmed
:> > with other compilers too, the special handling might also be removed
:> > from unexelfsgi.c, or we might remain with unexelf.c, even on IRIX6.
:>
:> Beware of sgi's compiler. Really.
SGI compiler is ok, SGI's rld is a bit more involved, thou, especially
if rqs or cord were around. So, to be on the safe side, I'd say - keep
the stuff as it is, don't move it. After all, we don't need nothing in
the sections which are not in .bss or .sbss, so just leave them as
they are.
I have done test builds of XEmacs21.1.8cvs+Max's patch on two systems:
IRIX64 ruby 6.2 06101031 IP28 mips with gcc 2.7.2.3
IRIX truth 6.5 01221553 IP32 mips with MIPSpro 7.2.1
executive summary:
everything works with the MIPSpro compiler but not with gcc.
gory details:
- the binary built on truth(6.5) works on ruby(6.2)
- the gcc build has the old, well-known problem that temacs has an
sbss section that is apparently not handled correctly by
unexelfsgi.c (despite ChangeLog entries implying this)
- I can get rid of the sbss section by adding '-G0' to the LDFLAGS
and then get a perfectly fine xemacs with gcc. Actually all my
production XEmacsen (21.1.4) were built this way on IRIX5.3.
I have put the Installation files and the output of 'objdump
--section-headers' at
http://svr-www.eng.cam.ac.uk/~ge204/XEmacs/
Let me know if you want me to do more tests, whether all this is
considered normal behaviour, where to start debugging, etc. I can also
test on IRIX5.3 if necessary...
Gunnar