Re: Testing MinGW Octave-3.4.2 binaries

[Philip Nienhuis]

Philip Nienhuis-3 wrote:
> 
> Jordi Gutiérrez Hermoso wrote:
>> On 7 July 2011 17:13, Philip Nienhuis<address@hidden>  wrote:
>>
>>> 1. "clear all" leads to a complete crash.
>>
>> Under all circumstances?
> 
> Always.
> 

No, I was wrong: when all traces of the windows package are gone (i.e.,
unload it, then uninstall it and only then restart Octave) "clear all" works
OK. 
Just a "pkg unload windows" is not sufficient.

Using a debug build, I found that the crash happens in symtab.h, when
clear_cmdline_function.islocked() is called. Commenting that out &
recompiling & again running in gdb, I see that the crash (signal) now occurs
at the next statement.
That makes me suspicious that something is clobbered up before the call to
clear_cmdline_function(). Unfortunately I have little experience with gdb,
so this will take some time to sort out.

BTW  I made a debug build to sort out other issues with the windows package.
So this issue actually belongs in the Oct-Dev mailing list (where I already
have a thread about this).




>>> 2.  A Matlab script, made by a colleague, that I'm porting to Octave
>>> creates 34 pictures. On some PC's, using fltk, doing "close all"
>>> leads to a similar crash as 1. (above)
>>
>> This is a long standing race condition with fltk, and I think it's the
>> biggest thing we have to fix before we switch the default plotting
>> engine from gnuplot to fltk. Known problem, also happens outside of
>> Windows.
> <snip>
> 

It can't reproduce this anymore. I can't find the cause.
As this is at work, I have little time investigating.
So just forget about this.



>>> 3. With that script, again on some PCs only, using fltk, titles
>>> appear only on the first plot. Using gnuplot all plots get the
>>> proper titles.
>>>
>>> 4. Again, on some PCs only, text output in fltk pics is gibberish.
>>> May be due to lacking fonts - I haven't had time to investigate.
>>> Gnuplot plots are fine.
>>
>> These two may be again more UB symptoms of the same race condition.
>> Does the stupid workaround described above also work on these?
> 
> I'll check at work tomorrow (if I've got time).
> 

Regarding the fonts (issue 3): This is a matter of default font setting in
Octave that seems to be wrong for our work PCs. I don't know how to adapt
this from the Octave side.
As this is on a fairly locked down office PC from my employer I can't do
much about fonts stuff from that side.

Issue 4: Can't reproduce it anymore.
In hindsight I think my work PC was a bit overloaded when I tried - IIRC the
mail system crashed at the end of the day as well.
So let's forget about this one as well.


All in all I think Tatsuro's binary works quite good for me. The only issues
left are #3 and #5 in my original post (edit.m). AFAICS the workaround for
#5 (using double backslash in the EDITOR string) is "stable".

In Benjamin's version 3.2.4-MinGW there were issues with the sse3 atlas lib
with complex arithmetic. Hopefully this is solved now.


Philip

--
View this message in context: 
http://octave.1599824.n4.nabble.com/Testing-MinGW-Octave-3-4-2-binaries-tp3652644p3659682.html
Sent from the Octave - Maintainers mailing list archive at Nabble.com.