Re: 3.4/3.5 PATCH: Fix IRIX 5 bootstrap

[Richard Sandiford <rsandifo at redhat dot com>]

Rainer Orth <ro@TechFak.Uni-Bielefeld.DE> writes:
> * irix_asm_named_section wasn't defined for IRIX 5, giving compiler
>   warnings-turned-to-errors.  The current version handles all ABIs and both
>   SGI and GNU as, so the TARGET_ASM_NAMED_SECTION definition can be moved
>   from iris6.h to iris5.h and the iris5gas.h variant removed.

This part is OK, thanks.

> * Definitions of current_section_name and current_section_flags were
>   missing.  We cannot simply move the iris6.h definition to iris5.h since
>   it doesn't compile: in_bss isn't defined without BSS_SECTION_ASM_OP.
>
>   I have the suspicion that the current iris6.h hack to use .data for the
>   O32 BSS_SECTION_ASM_OP causes some g77 testsuite failures with the native
>   as, so I don't want to go for the same hack here, but rather duplicate a
>   stripped-down version in iris5.h.
>
>   This is ugly, but so is the whole EXTRA_SECTION_FUNCTIONS stuff where you
>   can't test e.g. for BSS_SECTION_ASM_OP before using in_bss.

Hmm.  From a quick look at varasm.c, I couldn't see how defining
BSS_SECTION_ASM_OP (and only defining BSS_SECTION_ASM_OP) would change
the way gcc deals with the no_named_sections case.  Does it really have
an effect?

If http://gcc.gnu.org/ml/gcc-patches/2003-09/msg01064.html
is anything to go by, the ld problem is caused by a label at the
end of the data section.  Specifically the label passed to s_stop().
Where is this label placed for IRIX 5 after your current patch?

Richard