[Gc] Fwd: [boehm-gc] Use __rld_obj_head for IRIX GC_register_dynamic_libraries (PR libgcj/49451)

[Rainer Orth]

Hans,

> I expect it is.  Unfortunately, I think it requires a bit of rework by
> someone with access to an Irix (and ideally Solaris) machine, to make
> it fit there.

I could do that, on the other hand I doubt many people are using
mainline boehm-gc on IRIX anyway :-)

> The real cause of the problem here seems to be that we can't use a
> cached file descriptor, since (according to the comment) some kernels
> get upset if we have an open /proc fd when the process is killed.  Is
> that still an issue?  Does just removing the last two lines of that
> version of GC_register_dynamic_libraries(), perhaps just on Irix,
> still break things?

I have no idea to which platforms this comment applies.  This would
probably require some code (or mailing list) archeology.  One advantage
of the new __rld_obj_head code is that it doesn't require any system
resources (fds, system calls) at all.

> I'm OK with the patch as is (and I'm impressed that Rainer managed to
> uncover the right information), but it seems to add complexity and an
> ugly implementation-dependency for somewhat of a corner case on a
> somewhat obsolete OS.  (The Wikipedia entry starts with "Irix was" :-)
> ) If we could fix it by deleting code instead ...

Says Hans who worked for SGI for a long time :-)  But I agree: I'm
currently making sort of a final pass over the remaining IRIX (and Tru64
UNIX) testsuite failures and bugs, and will probably obsolete both
platforms for GCC 4.7, to remove all remnants in 4.8.

I think the best way forward is to apply the patch to the gcc tree only
without a forward port to gc mainline.  It seems unlikely that the
in-tree version will be updated soon, so I'll consider that only if this
happens.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University