[MMTK] netcdf on linux - undefined symbol: _fxstat
peter at maubp.freeserve.co.uk
Thu Jul 15 15:46:02 CEST 2004
hinsen at cnrs-orleans.fr wrote:
> On 14.07.2004, at 15:38, Peter wrote:
>> The Suse linux machine I'm trying to install MMTK already has Python
>> 2.3 installed, and I'm trying to install Scientific Python 2.4.6 and
>> MMTK 2.4.1 under my home directory. This seems to have worked,
>> except for netcdf:
> Isn't there a precompiled netCDF on the SuSE CDs? There was for version
> 8.1 (which I am using), so I suppose it is also in yours, which must be
> newer if it has Python 2.3.
There is in fact a compiled netCDF already installed:
(Note that the .a lib files are not marked as executable - but then
hardly any of the other lib files in /usr/lib/ seem to be either)
Doing "rpm -q netcdf" tells me its "netcdf-3.4-4", while the version I
had downloaded, compiled and installed under my home directory was 3.5.1
Anyway, I just removed my locally installed python modules and netCDF,
and rebuilt Numeric 23.1 (NumPy) then Scientific Python 2.4.6.
Scientific Python does seem to pickup this system copy of netCDF, but I
still get the same _fxstat error as before:
>> Trying "import Scientific.IO.NetCDF" gives this error:
>> Traceback (most recent call last):
>> File "<stdin>", line 1, in ?
>> ImportError: ./Scientific_netcdf.so: undefined symbol: _fxstat
>> My googling would appear to suggest _fxstat (or at least fxstat) is
>> handled by a shared libc file...
> Yes, it is part of the standard C library that every executable is
> linked to. Python modules "inherit" this from the Python interpreter.
> So is you find no other solution (I have no idea where this comes from,
> or rather, lots of vague possibilities), you could try to build your
> personal Python distribution in your home directory and then add NumPy,
> ScientificPython, and MMTK.
Might try this then...
Does the environment variable LD_LIBRARY_PATH sound relevant? Its
currently set to this:
More information about the mmtk