This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: xfail gcc.dg test


On Sat, Oct 05, 2002 at 03:00:40PM -0700, Richard Henderson wrote:
> On Sat, Oct 05, 2002 at 08:30:19PM +0930, Alan Modra wrote:
> > No, I'm not renaming the .text *section*, just sneakily renaming the
> > .text section *symbol*.  Same for .data, .bss etc.
> 
> Please don't.  At present, one can use the section symbol and the
> section name interchangably.  For _any_ section, not just those
> predefined by the gABI.  Changing this would be extremely confusing.
> Much much worse than differing the code symbol between xcoff and elf.

How often do you reference a section symbol?  Not very often.  And I'm
proposing the change for all sections, not just predefined ones.  Thus
"some_arbitrary_section" has ".some_arbitrary_section" as its section
symbol on powerpc64-linux.  There's even some consistency with branch
and link to ".foo" when you want to call function "foo".

Hmm, a better idea.  If we are going to change the ABI, then it would
be nice to get rid of those dot symbols entirely, replacing them with
a local symbol.  The linker already translates "bl .foo" when linking
dynamic objects, to a stub that uses the opd entry for foo.  We could
do a similar trick when linking statically, ie. "bl foo" (new scheme
would reference the opd sym) gets translated to "bl <address in opd
entry>".  The impact on the toolchain will only be marginally more
than the change to use "..foo", and removes our major kludge.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]