core dumped after importing matplotlib on SunOS

Community matplotlib-users
1

Hello, I've juste downloaded the last version (0.8) of

    > matplotlib and successfully built it on SunOS 5.8 with GTK
    > backend (gcc 3.2.2). "import matplolib" is OK but typing
    > "from matplotlib.pylab import *" result in a core dumped !
    > without any error message or exception... "import gtk",
    > "import pygtk", and "import wx" works OK.

    > I've tried with version 0.72, same error :

I don't have a lot of good advice for you, but here is how I would go
about debugging the problem. You can recompile mpl in verbose mode.
rm -rf "site-packages/matplotlib" and your "build" subdir and edit
setup.py and set VERBOSE=True and recompile/reinstall.

This can sometimes provide information about which extension code
function is failing.

You also want to create a test script that replicates the problem (in
your case it's a one-liner that simply imports matplotlib.pylab) and
run it with the --verbose-debug flag

   > python myscript.py --verbose-debug

This prints maximal debug info which can provide helpful information
to narrow down the possible culprits.

    > Python 2.4.1 (#3, Apr 14 2005, 10:58:02) [GCC 3.2.2] on
    > sunos5 Type "help", "copyright", "credits" or "license" for
    > more information.
    >>>> from matplotlib.pylab import *
    > Bus Error (core dumped)

The old-fashioned but effective way to triangulate this kind of
segfault is to put a lot of print statements into the module that is
segfaulting (in this case site-packages/matplotlib/pylab.py). Space
these print statements every 50 lines or so

  print 'debug 1'
  ...50 lines of pylab...
  print 'debug 2'
  ...50 lines of pylab...
  print 'debug 3'

When you have narrowed the segfault to be between two numbers, say 2
and 3, add more statements between them, maybe with a delimiter such
as TAB '\t'

  print '\t debug 2a'
  ...5 lines of pylab...
  print '\t debug 2b'
  ...5 lines of pylab...
  print '\t debug 2c'
  ...5 lines of pylab...

and repeat as necessary with a finer grain until you find the actual
line causing the segfault. Posting this info will go a long way to
actually discovering the bug.

I call this "poor man's debugging" or "manual binary search" and it's
primitive, ugly, and highly effective.

    > One more thing is that after importing matplotlib, I got an
    > error when using help :

Don't know if this is related or not...

JDH

2

    > Hello, I've juste downloaded the last version (0.8) of
    > matplotlib and successfully built it on SunOS 5.8 with GTK
    > backend (gcc 3.2.2). "import matplolib" is OK but typing
    > "from matplotlib.pylab import *" result in a core dumped !
    > without any error message or exception... "import gtk",
    > "import pygtk", and "import wx" works OK.

    > I've tried with version 0.72, same error :

I don't have a lot of good advice for you, but here is how I would go
about debugging the problem. You can recompile mpl in verbose mode.
rm -rf "site-packages/matplotlib" and your "build" subdir and edit
setup.py and set VERBOSE=True and recompile/reinstall.

Ah ! that was what I was looking for !
So I've managed to isolate the culprit :

from font_manager import FontProperties

Glyph::init_type
FT2Font::init_type
ft2font_module::new_ft2font
FT2Font::FT2Font
FT2Font::clear
Illegal Instruction (core dumped)

It appears that the problem was in file ft2font.cpp at line 354 :
const char* ps_name = FT_Get_Postscript_Name( face );

I don't know what is wrong here but I managed to get matplotlib working
by replacing it by :
const char* ps_name = NULL;

I forgot to take the version number of the freetype library before
leaving work, I will get it on monday.

I haven't understand what is really in the argument 'face', can you give
me a code which print the fields of 'face' ? (I don't know C++...)

Thank you anyway for your help !

···

On Fri, 15 Apr 2005 00:34:15 -0500, John Hunter <jdhunter@...4...> wrote :

3

:

I forgot to take the version number of the freetype library before
leaving work, I will get it on monday.

Version of freetype is 2.1.7 !

···

On Sat, 16 Apr 2005 00:19:37 +0200, Olive <enchere_email@...136...> wrote

4

Hi Olive,

I've had similar problems on both Solaris 8 and 9. By running python under truss:

   truss python

I saw that the "illegal instruction" and core dump was happening in the freetype libraries. I had built freetype 2.1.9 myself. When I deleted my own libfreetype.* and instead used the prebuilt versions /usr/sfw/lib/libfreetype.* my crashes went away.

(Note that /usr/sfw/lib is the Solaris 9 location. I'm not sure if it's the Solaris 8 location or not).

I had built libfreetype.6.3.7 (freetype 2.1.9) and /usr/sfw/lib contained the earlier version libfreetype.6.3.1 . I don't whether this is a Solaris-only bug in later versions of freetype, or if the /usr/sfw/lib versions were built a different way than I built mine.

Anyway, that's how I solved the problem.

I would suggest that you try:

(1) Finding the Sun-supplied libfreetype.so binaries and using those, or
(2) building the version of libfreetype that produces libfreetype.6.3.1.

Please let me know how you solve the problem.

Good luck!

--Michael Brady

5

Hello Michael,

Like I said I solved temporarily the problem by editing the file
ft2font.cpp (from matplotlib !) at line 354 :
const char* ps_name = FT_Get_Postscript_Name(face );

I replaced it by :
const char* ps_name = NULL;

It's just a quick and dirty hack while waiting for a clean patch from
either matplotlib or freetype.
I still don't know wether it's a bug from matplotlib or freetype,
because I don't know what is the argument 'face' and what is inside...
I don't understand a thing in C++ so I can't be of much help here...

But I was working with freetype 2.1.7, I have also tried with 2.1.9 and
I also have a core dumped but no at the same place !
Don't know if my hack could work also with 2.1.9 ?
I can test if you're interested.

P.S. : I couldn't help noticing your email address, I'm also working on
space rockets but in France and also trying to promote matplotlib here !
:wink:

···

On Thu, 21 Apr 2005 08:27:41 -0700 (PDT), Michael Brady <mbrady@...369...> wrote :

Hi Olive,

I've had similar problems on both Solaris 8 and 9. By running python
under truss:

   truss python

I saw that the "illegal instruction" and core dump was happening in
the freetype libraries. I had built freetype 2.1.9 myself. When I
deleted my own libfreetype.* and instead used the prebuilt versions
/usr/sfw/lib/libfreetype.* my crashes went away.

(Note that /usr/sfw/lib is the Solaris 9 location. I'm not sure if
it's the Solaris 8 location or not).

I had built libfreetype.6.3.7 (freetype 2.1.9) and /usr/sfw/lib
contained the earlier version libfreetype.6.3.1 . I don't whether this
is a Solaris-only bug in later versions of freetype, or if the
/usr/sfw/lib versions were built a different way than I built mine.

Anyway, that's how I solved the problem.

I would suggest that you try:

(1) Finding the Sun-supplied libfreetype.so binaries and using those,
or(2) building the version of libfreetype that produces
libfreetype.6.3.1.

Please let me know how you solve the problem.

Good luck!

6

Is it currently possible to make contour plots using unstructured data with matplotlib?

By unstructured data I mean the typical data from a 2D finite element simulation that
uses an unstructured polygonal mesh.

As far as I can tell the "pylab.contourf(X, Y, Z)" command requires structured data with X, Y
and Z each having to be 2D gridded arrays.

I have previously used the gist package to do this type of plotting using the
"gist.plfp(Z, X, Y, F)" command. This command allows plotting of polygonal data
specified by arrays Z, X, Y, F.

     - Z is the data value, shape(Z) = (NPolygons)
     - X is the vertex x coordinates, shape(X) = (NPolygons, NVertices).
     - Y is vertex y coordinates, shape(Y) = (NPolygons, NVertices)
     - F is the number of vertices for each cell, shape(F) = (NPolygons)

Is there a similar functionality in matplotlib?

Thanks

···

-------------------------------------
Daniel Wheeler