I've been having a lot of problems with intermittent crashes while
using RPy under a rather old Python 2.4.2 install, with no Numeric
support whatsoever. This is all under Linux.
Here's an example of the code which causes a crash. I'm loading an R
file and want to verify that it only contains a single object and
that I haven't seen those object names before.
_already_loaded_names = {}
def load_clab_from_rdata_file(filename):
from rpy import r
name = r.load(filename)
if isinstance(name, list):
raise TypeError("R model %r defines more than one object: %r" %
(filename, name))
if name in _already_loaded_names:
raise TypeError("%s from %r already loaded from %r" %
(name, filename, _already_loaded_names[name]))
_already_loaded_names[name] = filename
return r[name]["clab"]
There's nothing complex here. When I try it out (the TEST_r_simple2
directory contains the file 'rmodel.RData', which has two objects) I
get this sort of error.
% ./drone_properties.py --smiles c1ccccc1O TEST_R_simple2 TEST_R_simple
Error when computing 'TEST_R_simple2' of 'c1ccccc1O'
Traceback (most recent call last):
File "Properties.py", line 211, in resolve_named_properties
props[name]
File "Properties.py", line 101, in __getitem__
handler(key, self)
File "/ModelBase.py", line 138, in make_single_prediction
self.make_prediction([dict])
File "RModel.py", line 98, in make_prediction
self.predict(cmpd)
File "RModel.py", line 121, in predict
self._predict = load_clab_from_rdata_file(
File "RModel.py", line 59, in load_clab_from_rdata_file
raise TypeError("R model %r defines more than one object: %r" %
TypeError: R model 'TEST_R_simple2/rmodel.RData' defines more than
one object: ['a', 'b']
*** caught segfault ***
address 0x34b100, cause 'memory not mapped'
Possible actions:
1: abort (with core dump, if enabled)
2: normal R exit
3: exit R without saving workspace
4: exit R saving workspace
Selection: 1
Exception thread.error: 'release unlocked lock' in <class
thread.error at 0xb63083bc> ignored
I can generate a core dump but it's in some completely unrelated
shared library that's never given problems before. There are some
several dozen shared libraries in the system.
I'm not able to create a minimal test case.
I tried running it under valgrind's memtest but found nothing out of
the ordinary. The exception did occur, but valgrind reported nothing.
What's interesting is if I force all malloc'ed data to 0 then the
problem disappeared
% valgrind --malloc-fill=00 --leak-check=yes ./drone_properties.py --
smiles c1ccccc1O TEST_R_simple2 TEST_R_simple
It didn't reappear if I used FF or 23, so I don't know if this is
coincidence.
Looking through the archives I've found no mention of this problem.
Any ideas? Is my version of Python too old, with a since resolved
bug? Or is it there some nasty interaction between the Rpy thread
and all of the system/popen/etc. calls I do elsewhere in the software?
How do I turn off R's override of the signal handlers so I can get a
more normal behavior that something like valgrind can help me out with?
If I can't get this working, I think I saw a popen-based interface to
calling R, which I would like to try out. Does anyone know where it is?
Best regards,
Andrew
[EMAIL PROTECTED]
-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
rpy-list mailing list
rpy-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/rpy-list
