3.2 ia64: conflict with asms and -g

["Wichmann, Mats D" <mats dot d dot wichmann at intel dot com>]

Not sure if this is bug or usage problem - so I'm
looking for advice from someone who knows, particularly,
the ia64 port since it seems to be unique to that one.

A project I'm on has a set of stub libraries generated 
from a database which are used at link time to control
symbol inclusion; normal libraries are used at runtime.

The stubs are normally built with "-g". When building 
with gcc 3.2, a few of these fail this way, but build 
if "-g" is omitted.  2.9x gcc does not seem to have 
this issue, nor does 3.x on ia32 or ppc.

This fragment reproduces the problem for me:

void acos(void) {} ;
__asm__(".globl signgam; .data; .type signgam,@object; .size signgam, 4;
signgam: .long 0");


When this is built to assembler with a 2.96 compiler,
we see something like this:

<snip>
#APP
        .globl signgam; .data; .type signgam,@object; .size signgam, 4;
signgam: .long 0
 
        .section        .text
.Letext0:
 
        .section        .debug_abbrev, "", "progbits"
</snip>
<snip>
	data8.ua	.Letext0-.Ltext0
</snip>


The same bit with a 3.2 compiler looks like this:

<snip>
#APP
	.globl signgam; .data; .type signgam,@object; .size signgam, 4;
signgam: .long 0
.Letext0:
	.section	.debug_info
	data4.ua	0x45
	data2.ua	0x2
	data4.ua	@secrel(.Ldebug_abbrev0)
</snip>
<snip>
	data8.ua	.Letext0-.Ltext0
</snip>

The .section .text statement is missing from the latter.
In other words, .Letext0 looks like it's now in
data, rather than text, and the assembler gags on
the calculation ".Lextext0-.Ltext0".  This isn't 
terribly hard to work around; the script that generates 
the stub can add a ".text" at the end of each asm line 
that has a ".data".  

What I'd like to find out is if this is a bug in
gcc, or if it's not considered kosher to use
section declaration in asm statements, or what
the story is.

Thanks much.

-- mats wichmann