Re: Odd crashes with Java

[Rik]

On 12/16/2012 04:56 AM, address@hidden wrote:
> Message: 9
> Date: Sun, 16 Dec 2012 04:55:58 -0800 (PST)
> From: Philip Nienhuis <address@hidden>
> To: address@hidden
> Subject: Re: make check: segmentation fault in triplequad.m (in CLI)
> Message-ID: <address@hidden>
> Content-Type: text/plain; charset=us-ascii
>
> Philip Nienhuis wrote
>> > 15794:cf3bb2f353a5 tip
>> > Linux Mageia 2
>> > 
>> > address@hidden oct370]$ make check
>> > 
>> > :
>> >   libinterp/interpfcn/dirfns.cc .......................... PASS    1/1   
>> >   libinterp/interpfcn/file-io.cc ......................... PASS    0/1   
>> > FAIL 1
>> >   libinterp/interpfcn/graphics.cc ........................ PASS   14/14  
>> > :
>> > (the usual FAIL for file-io.cc, locale issue. There was a new FAIL for
>> > logspace, I've fixed that)
>> > :
>> >   scripts/general/trapz.m ................................ PASS   11/11
>> >   scripts/general/triplequad.m ...........................make[1]: ***
>> > [check] Segmentation fault
>> > make[1]: Leaving directory `/home/philip/devel/oct370/test'
>> > make: *** [check] Error 2
>> > address@hidden oct370]$
>> > 
>> > Interestingly, in the CLI (./run-octave --no-gui):
>> > 
>> > octave:6> test triplequad
>> > Segmentation fault
>> > 
>> > ...while in the GUI:
>> > 
>> > octave:1> test triplequad
>> > PASSES 1 out of 1 tests
>> > octave:2>
>> > 
>> > Octave (GUI) also seems unstable (I got several hangs).
>> > 
>> > Could this also be related to bug #37924 ?
>> > 
>> > Philip
> Answering to myself:
>
> Apparently so. After having added some tests to ov_java.cc and rebuilding,
> make check now chooses to crash at:
>
> :
>  scripts/general/curl.m ................................. PASS    1/1
>   scripts/general/dblquad.m ..............................make[1]: ***
> [check] Segmentation fault
> make[1]: Leaving directory `/home/philip/devel/oct370/test'
> make: *** [check] Error 2
>
> Philip
12/16/12

I think we do have some sort of instability that was introduced with Java. 
I discern this particularly because the crashes move around when the code
is re-compiled.  Besides bug #37924, there is #37891 about a crash in
contourc, and #37877 about crashes when using __java_init__ and __java_exit__.

Does 'make check' pass repeatedly if you configure with '--disable-java' ?

--Rik