From mboxrd@z Thu Jan  1 00:00:00 1970
From: Ian Lance Taylor <ian@cygnus.com>
To: don@pixsci.com
Cc: gas2@cygnus.com
Subject: Re: Patch for binutils (GAS-980329) for VxWorks on MIPS
Date: Mon, 30 Mar 1998 08:34:00 -0000
Message-id: <199803301634.LAA23997@subrogation.cygnus.com>
References: <3521B963898BD111AA6A006008A84516079B37@SERVER>
X-SW-Source: 1998/msg00073.html

   From: Don Bowman <don@pixsci.com>
   Date: Sun, 29 Mar 1998 18:46:03 -0500

   Under VxWorks 5.3.1 [The Tornado 1.0.1 release] for the MIPS targets,
   gas cannot create a debuggable image. This occurs since the Tornado
   target server requires a separate .stab & .stabstr section. It cannot
   deal with the .mdebug ecoff section. 
   The fix is to three main areas:
   1: configure: accept 'mips-wrs-vxworks' as a valid configuration target
   2: obj-elf.h: allow output to separate sections. The comment that was
      there indicated this had been removed due to an Irix 6 limitation. This
      may not affect any of the other ELF targets, but I cannot say for
   certain:
      it certainly does not affect the VxWorks MIPS platform.

The .mdebug section is required on Irix 5, because the Irix 5 linker
can not handle .stab/.stabstr sections.  I do not believe the .mdebug
section is required on Irix 6, although gas doesn't support Irix 6
anyhow so I haven't tested this.

It appears that your patch will always generate .stab/.stabstr
sections, even on Irix 5, since MIPS_STABS_ELF is always defined in
obj-elf.h in your patch.

Presumably WRS ships a toolchain which permits debugging for MIPS
targets.  What do they do?

   3: A fix to ecoff.c: the comment about input_line_pointer pointing to the
      beginning of the stab value appears incorrect.

This appears unrelated to the other issues.  Can you send a test case
for this?  As far as I can tell from s_stab_generic,
input_line_pointer will indeed point to the start of the value when
ecoff_stab is called.  Without a test case, I can't judge whether your
patch is appropriate.

Ian