Re: ia64 libjava java-signal.h build failure

[Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>]

In article <or8zkuhtmr.fsf@guarana.lsd.ic.unicamp.br> Alexandre Oliva writes:

> But if we don't enable GCJ by default, it's that logic that won't get
> tested, unless people configure --enable-languages=foo,bar,java.  I've
> already done it on a number of platforms for the past 2-3 weeks, and
> reported that I've encountered problems only on FreeBSD 2 (because of
> a local change that caused the generic FreeBSD code to disable libjava
> not to work) and AIX 4.1 (actually, on GNU/Linux/sparc too, but that
> problem is already fixed).  On all other platforms, namely
> GNU/Linux/x86, GNU/Linux/alpha, DU4.0D, Solaris 7/x86 and /sparc,
> Solaris 2.6 and 2.5/sparc, FreeBSD 3.0, 3.4 and 4.0/x86, IRIX 6.3 and
> 5.2 and HP-UX 10.20, a Java-enabled build just worked out of the box.

Hi Alexandre, et al.,

I do not dispute that GCJ builds on recent FreeBSD by default
(i.e. without any configure-time arguments) since I have no
information but for the record:

Whenever FSF gcc is installed on a FreeBSD system, it must be
configured with --enable-threads\=posix or else you lose features
compared to the system compiler.  Unless something changed in the last
few days, GCJ does not build under this configuration.  I have looked
at the issue but unfortunately I have concluded that the size of the
patch is well beyond what I am allowed to contribute as a "bug fix"
without official paperwork.

The main problems are assumptions on how threading libraries are
named/found within the GCJ code base (FreeBSD 5 will be
POSIX-compliant or at least Linux-compliant on this point, but the
releases named above are not) and that the version of the Boehm
collector checked into the gcc tree only knows how to do POSIX threads
on specific OSes.

Until GCJ builds with the top-level configure-time argument
--enable-threads\=posix, please do not enable it by default on
FreeBSD.

Regards,
Loren