Re: [PATCH] Fix libjava breakage with glibc 2.2.2pre

[Bryce McKinlay <bryce at albatross dot co dot nz>]

Jakub Jelinek wrote:

> The bug in linuxthreads is fixed since 1999-08-19 (and actually now
> __sigaction is defined in -lpthread as well, so if it would not be fixed,
> the trick with __sigaction would not work either).
> So I guess there are two options: either use syscall(SYS_sigaction)
> everywhere instead of __sigaction, or a patch below.
> This should be safe, because the linuxthreads bug is fixed in glibc-2.1.3
> and above and if one compiles libgcj on a box with glibc-2.1.3 and above, it
> uses at least some symbols @GLIBC_2.1.3 (like __cxa_finalize), so one cannot
> use that libgcj DSO under older (buggy in this regard) glibcs anyway.

I believe that this bug (or a similar one) is still present, because it was just a few weeks
ago that Andrew put in the syscall(SYS_sigaction) to fix a problem with the divide handler
that was showing up on glibc {2.1.3,2.2.1}/kernel 2.4. You can see a description of that
problem here: http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=1718&database=gcc

Ideally I would like to see glibc/linux fixed, especially given that using the syscall
directly may break any possible future M:N linuxthreads implementation, but for now I
suspect that it may be the only working solution.

regards

  [ bryce ]