Re: [PATCH] PLT reducation, take 2

[Jakub Jelinek <jakub at redhat dot com>]

On Sat, Aug 03, 2002 at 12:09:45AM -0700, Ulrich Drepper wrote:
> Jakub Jelinek wrote:
> 
> > I've added the configure check plus a bunch of new macros from PLT-reduce
> > list and converted a bunch from INTUSE/INTDEF etc. too.
> > Tested by make check, comparing nm -D output and readelf -r output before
> > and after the patch.
> 
> I have applied the patch now.  But while doing this I had to add a few 
> more changes.  There are more gcc problems which are made visible by 
> this change, not only the visibility+asm problem (see the new test in 
> configure.in).
> 
> The result is new compilable again and it seems to work
> 
> What I don't see is the PLT reduction.  I might have broken something by 
> getting things to compile or there are more tools problems.  This has to 
> be investigated.  The design is clean and it compiles which is the most 
> important thing.  Please try it with your compiler to see whether there 
> are yet more problems.

Well, with gcc-3.2-0.1.1 (ie. the compiler which no longer outputs
.hidden *foo
if visibility hidden is mixed with __REDIRECT (that patch is in CVS head too)),
I got following:

immediately before the patch:
readelf -a libc.so | grep '\(^Relocation\|RELCOUNT\)' 0x6ffffffa (RELCOUNT)                   1406
Relocation section '.rel.dyn' at offset 0x11798 contains 1525 entries:
Relocation section '.rel.plt' at offset 0x14740 contains 301 entries:

with the patch:
readelf -a libc.so | grep '\(^Relocation\|RELCOUNT\)' 0x6ffffffa (RELCOUNT)                   1405
Relocation section '.rel.dyn' at offset 0x11798 contains 1522 entries:
Relocation section '.rel.plt' at offset 0x14728 contains 270 entries:

	Jakub