From mboxrd@z Thu Jan  1 00:00:00 1970
From: Ian Lance Taylor <ian@zembu.com>
To: nickc@cygnus.co.uk
Cc: binutils@sourceware.cygnus.com
Subject: Re: Patch: Support .loc STABS directive on ELF based toolchains.
Date: Wed, 29 Sep 1999 08:01:00 -0000
Message-id: <19990929150032.26734.qmail@daffy.airs.com>
References: <199909291437.PAA30604@pathia.cygnus.co.uk>
X-SW-Source: 1999-09/msg00204.html

   Date: Wed, 29 Sep 1999 15:37:42 +0100
   From: Nick Clifton <nickc@cygnus.com>

     Here is a small patch to allow the .loc STABS directive to be used
     on ELF toolchains which use .gnu.linkonce.t sections as special
     cases of the .text section.

   1999-09-29  Nick Clifton  <nickc@cygnus.com>

	   * ecoff.c (ecoff_directive_loc): Allow the .loc directive to
	   appear inside .gnu.linkoce.t. sections for ELF based
	   toolchains. 

We don't reject .loc outside of a .text section because we are
mean-hearted.  We reject it because it won't work.  ECOFF debugging
information builds a single table of line numbers for an object file.
That table is implicitly assumed to be attached to the .text section.
I don't see how to avoid that, given the weird construction of the
ECOFF symbol table.

If you want to use ECOFF debugging information, I think you need to
avoid using linkonce sections.  Or vice-versa.  I recommend using
straight stabs debugging information instead.

If the debugging information really does work correctly, particularly
after processing through ld -r, let me know (by work correctly, I mean
that objdump -dl and gdb can correctly report line numbers for the
linkonce functions in question).

   !       const char name[] = ".gnu.linkonce.t.";

I don't think K&R C compilers support automatic array initialization.
Make it static.

Ian