python run_all.py segfaults Mac OS X 10.5.6

[dopplershift]

From watu...@gmail.com on March 03, 2009 07:25:07

What steps will reproduce the problem? 1. Follow documentation to install netCDF4-0.7.7 and all required packages (zlib, hdf5, netcdf-
4, all of which test correctly)

1. Do rudimentray test:

hank:test tjs$ python
Python 2.5.1 ( r251 :54863, Jan 13 2009, 10:26:13)
[GCC 4.0.1 (Apple Inc. build 5465)] on darwin
Type "help", "copyright", "credits" or "license" for more information.

import netCDF4

1. cd to test directory of netCDF4 src distribution and python run_all.py What is the expected output? What do you see instead? Expected: Not a seg fault, I presume.

hank:test tjs$ python run_all.py
..E.Segmentation fault

This happens in about 2 seconds

hank:test tjs$ python --version
Python 2.5.1
hank:test tjs$ python
Python 2.5.1 ( r251 :54863, Jan 13 2009, 10:26:13)
[GCC 4.0.1 (Apple Inc. build 5465)] on darwin
Type "help", "copyright", "credits" or "license" for more information.

What version of the product are you using? On what operating system? netCDF4-0.7.7

MacOS X 10.5.6 Please provide any additional information below. quad-core mac pro

latest Xcode

hdf5-1.82

netcdf-4.0

zlib-1.2.3

ifort 11.059

gcc: i686-apple-darwin9-gcc-4.0.1

Original issue: http://code.google.com/p/netcdf4-python/issues/detail?id=13

[dopplershift]

From whitaker.jeffrey@gmail.com on March 03, 2009 07:50:29

Could you try updating netcdf-4 to the 4.0.1-beta2 and let me know if the problem
persists?

I'm running on a mac with hdf5-1.8.2 and netcdf-4.0.1-beta2 with no problems.

[dopplershift]

From watu...@gmail.com on March 03, 2009 08:50:13

I installed netcdf-4.0.1-b2 (it passed all tests in make check) then
rebuilt your module and installed it, then ran python run_all.py in
test. Failed with seg fault just as before:

hank:netCDF4-0.7.7 tjs$ cd test
hank:test tjs$ python run_all.py
..E.Segmentation fault

[dopplershift]

From whitaker.jeffrey@gmail.com on March 03, 2009 09:56:22

Can you

1. 'import netCDF4' in the interpreter without an error?
2. run the tests individually (do they all segfault)?
3. run the tests in the debugger (never have done this myself, but I found some hints
   at http://chrismiles.livejournal.com/20226.html ).

I'm a bit stumped - a virtually identical setup here (minus the intel compiler) works
fine. Did you use the intel c compiler to build either the libs or the python extension?

[dopplershift]

From watu...@gmail.com on March 03, 2009 10:22:54

1. Yes, that has worked from the start

2. tst_fancyslicing.py and tst_utils.py seg fault. The rest run OK.

3. I'll have to set up to do that. The gdbinit file doesn't seem to be on my system, so I'll need to find one
   appropriate for python 2.5.1

I don't use intel C because I don't have that. Using gcc 4.0.1 from Xcode 3.1.2.

What fortran compiler do you use?

What python version do you use?

[dopplershift]

From whitaker.jeffrey@gmail.com on March 03, 2009 10:29:47

OK, that helps. There is a known bug in the netcdf 4.0 lib that causes problems with
tst_fancyslicing.py and tst_utils.py. Are you sure you rebuilt the python extension
after updating to netcdf-4.0.1-beta2 (you may need to blow away the build directory
and rebuild)?

One other thought before you go to the trouble of running gdb - if you have a numpy
older than 1.2.1, could you update?

[dopplershift]

From watu...@gmail.com on March 03, 2009 10:42:12

I did blow away the netCDF4-0.7.7 directory after I built netcdf-4.0.1-b2. Then I untarred netCDF4-0.7.7.tar.gz
again and rebuilt from scratch (after updating environment variables to point at 4.0.1-b2).

Will do. I have Numpy 1.0.1. I assume I should rebuild your netCDF4 modules after updating Numpy, correct?

[dopplershift]

From whitaker.jeffrey@gmail.com on March 03, 2009 10:46:38

Yes, rebuild after updating numpy. I'll bet a beer that fixes it.

-Jeff

[dopplershift]

From watu...@gmail.com on March 03, 2009 11:45:48

Built and installed Numpy 1.2.1. Once I figured out how to make this the default and not the 4 other versions
still lingering, I rebuilt netCDF4-0.7.7, then ran test/run_all.py (after moving the dap test aside):

hank:test tjs$ python run_all.py

...............................

Ran 31 tests in 0.776s

OK

Eureka. Thanks for walking me through this.

I guess I owe you a beer.

[dopplershift]

From whitaker.jeffrey@gmail.com on May 25, 2009 05:59:14

Status: Verified