Re: sigcontext_struct

[Godmar Back <gback at cs dot utah dot edu>]

> 
> > From: Godmar Back <gback@cs.utah.edu>
> > Date: Mon, 18 Oct 1999 11:24:37 -0600 (MDT)
> > > 
> > > I think that we should drop support for the old sigcontext stuff
> > > altogether and use the newer SystemV siginfo handler instead.  I've
> > > already written this, and it's just a matter of checking it in to
> > > sourceware.  (Tom: it's in devo/libjava if you want to have a look.)
> > 
> > A bit of related information I thought I'd pass along.
> > 
> > As you know, siginfo doesn't work on Linux 2.0.x kernels.
> > However, there are some installations out there that run 2.0.x kernels
> > with versions of glibc that define SA_SIGINFO in their header files.
> 
> Yeah, but surely these are people who have an incompatible combination
> of library and kernel.  I can't see that libgcj is at fault if a
> header file is incorrectly defining a feature which is missing.
> 
> Actually, it seems odd that glibc is defining SA_SIGINFO at all; it's
> a kernel feature.
> 

I'm just reporting on what we've learned in Kaffe.  (I should credit
Edouard Parmelan specifically for finding this out.)

I know that major distributions were affected, so I don't think it
was individual people's fault.  Unfortunately, unlike in systems like
FreeBSD, if you upgrade a Linux kernel and/or glibc, there's no 
mechanism to ensure that the two are in sync.

Our work-around was to not use siginfo for the time being.
But you're right that it's slowly dying.

Some architectures don't even properly implement the old 
sigcontext.  Linux on the ARM is one such example where the kernel doesn't
properly establish the signal context.  But their implementation of
siginfo was broken too, the last time I checked.

	- Godmar