Re: [patch 3.4/3.5] libffi & libjava port for hppa-linux

[Andrew Haley <aph at redhat dot com>]

John David Anglin writes:
 > > Andrew Haley writes:
 > > > I guess I need to clarify what I said.
 > > > 
 > > > It's okay to check these changes into mainline, but until the test
 > > > failures are fixed we shouldn't enable Java by default.
 > 
 > That's unreasonable.  We would have to turn off support for all
 > languages on the PA on that basis ;(

Well, I was hoping to encourage someone to fix the few remaining bugs
first.

 > Users have control over what they build with --enable-languages.
 > 
 > > how is a language disabled for one specific target?
 > 
 > build_by_default could be set appropriately in java/config-lang.in.
 > 
 > > what are the criteria to disable it for one target, not for
 > > another (i.e. s390-linux shows 900 test failures in libjava).

Oh dear.  As far as I know it used to work, so the most likely
explanation is that there's a common fault that breaks everything.

 > Sparc and sparc64 have a comparable number of failures in 3.4
 > to that of the new PA support.  i686-pc-cygwin has more.
 > 
 > Enabling java by default will provide more testing and use, and
 > hopefully these problems will get fixed faster.

This is not my experience.  However, if there really are people
interested in using gcj on hppa-linux in its current semi-broken
state, then perhaps they can build with --enable-languages.  But at
the moment it doesn't look like we have a working
MD_FALLBACK_FRAME_STATE_FOR, and we really need that for proper
exception handling.

I suppose I'm mostly concerned about gcj's reputation as a decent Java
platform.  But if you really believe that enabling Java by default on
hppa-linux might get it working faster, I'll acquiesce.

Andrew.