Re: Basic block profiler out of libgcc

["Zack Weinberg" <zackw at Stanford dot EDU>]

On Sat, May 05, 2001 at 01:22:39AM +0200, Hans-Peter Nilsson wrote:
> 
> I see things have changed.  Now, if a target does not want a
> gcc-specific .ident, it has to be non-intuitive and define
> ASM_OUTPUT_IDENT but not IDENT_ASM_OP.  Before, it just had to
> provide an ASM_FILE_END that did *not* output an .ident for gcc
> (compared to config/elfos.h).  That does not seem like a good
> change.

I dunno, it seems to me that reducing by a large factor the number of
targets that need to dink with ASM_FILE_END is a good change.

I'm open to suggestions - perhaps we can find a way to make
ASM_OUTPUT_IDENT work for GCC's .ident, then we wouldn't need
IDENT_ASM_OP at all.

> Again (as I wrote in my patch to *remove* the .ident for gcc),
> the .ident for gcc is controlled by flag_no_ident but does not
> match the documentation for -fno-ident, which only mentions how
> #ident is treated; the documentation does not mention that it
> *always* emits one for gcc itself if this option is not used.

Patches are welcome.

> The .ident for gcc is not necessary and only has limited
> debug-related purposes, while accumulating *for each object* in
> executables.  It follows the same pattern as gcc2_compiled.  I
> think it should only be output with -g, if at all.

In my original patch I took it out completely.  A couple of people
pointed out uses for it which are independent of -g.  So I left it in.
The space issue should be solved by Jakub's string-merging patch
(automatically applied to .comment section, if I read it correctly).

-- 
zw                I'm on a spaceship full of college students.
                  	-- Martin "PCHammer" Rose