[parisc-linux] FYI: state of toolchain for Debian woody release

[parisc-linux] FYI: state of toolchain for Debian woody release

[Randolph Chung]

Just thought I'd post a quick note about some of the things happening
recently:

With many thanks to Dave Anglin and the investigation work by Carlos
O'Donell, we've finally fixed the bug in gcc-3.0 that affected passing
floating point arguments in indirect calls.

The new compilers are now in the Debian archive, and we are slowly
starting to rebuild some of the affected packages. (gs and python2.1
were the ones I knew about offhand, and they are now in the incoming
queue). If you know of any other Debian packages you use that might be
affected by this bug, let us know and we'll rebuild it.

There are still some known problems with the compilers, most notably the
problems with fixups that aren't able to reach their targets and thus 
requiring compilation with -ffunction-section. I have one patch from
Dave that fixes this in some cases, but as I understand it the general
problem is more involved and might not get fixed till post-woody.

gdb also seems to be rather broken on hppa at the moment. Single
stepping doesn't always (doesn't usually) work, and there's a problem 
with recognizing core files. If anyone has time to look into these 
and submit patches, that'll be great!

randolph
-- 
   @..@                                         http://www.TauSq.org/
  (----)
 ( >__< )
 ^^ ~~ ^^

Re: [parisc-linux] FYI: state of toolchain for Debian woody release

[John David Anglin]

> gdb also seems to be rather broken on hppa at the moment. Single
> stepping doesn't always (doesn't usually) work, and there's a problem 
> with recognizing core files. If anyone has time to look into these 
> and submit patches, that'll be great!

I haven't noticed any single stepping issues with the gdb that I built.
Stepping into functions doesn't work but single stepping does.  I have
made some changes to the dwarf2 frame note code in gcc and a working dwarf2
unwind mechanism is real close now.  I'm down to 4 fails in the g++ testsuite.

There are issues with loading symbols from shared libs.  They don't
load automatically even though they should.  You can't set new breakpoints
while a program is executing.  You have to run it again.  You can't
set breakpoints in shared libraries.

Also, can't print a function result.  This problem is probably related
to the stepping into function problem.

Info reg r3 prints the value of r3 saved in the frame and not the current
value in r3.  Not sure what happens if r3 is not saved in the frame.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)