This is the mail archive of the
egcs-testresults@egcs.cygnus.com
mailing list for the EGCS project.
Re: Results for gcc-2.95 19990618 (prerelease) testsuite on i686-pc-sco3.2v5.0.5
> > > eeprof1.c compilation (-fPIC, -melf)
> >
> > /usr/tmp/ccQmYq1b.s:212:name already bound as global: __cyg_profile_func_en
> > ter^M/usr/tmp/ccQmYq1b.s:226:name already bound as global: __cyg_profile_fu
> > nc_exit^M
> Losing assembler does not like ".globl" to appear more than once for
> a symbol. Note, we are _not_ changing the scope or type of the symbol,
> so I really think the assembler is barking unnecessarily in this case.
It is just a warning and not an error.
> This happens when you reference a libcall & define the libcall in the
> same compilation unit due to the ASM_OUTPUT_EXTERNAL_LIBCALL definition
> in sco5.h
I don't know offhand why 2.95 and 1.1.2 differ in this regard. Just
whacking the definition for ASM_OUTPUT_EXTERNAL_LIBCALL in sco5.h
allows eeprof to assemble warning-free. (I haven't run a full
bootstrap/testsuite.) Since that definition was blindly grafted in
from svr4.h and the comment in that file seems to imply that it was a
bug in some older assembler, perhaps we should just nuke it from modern
SVR4-ish assembler targets.
Since I seem to be wielding the scepter of waivers today, here's my
vote: Since it's merely a warning and not and error, if this only shows
up with -finstrument-functions, I can live with that. If whacking
the maco is an OK solution, I can go for that, too.
My vote could be swayed by bootstrap/testsuite results or by hints that
A_O_E_L is more important than it seems to be.
RJL