C lib segfaults when re-writing text attribute as string

[jswhit]

The following python script triggers the error:

from netCDF4 import Dataset
import numpy as np
nc = Dataset('test_att_convert.nc','w')
nc.foo = np.array('bar','S')
nc.close()
nc = Dataset('test_att_convert.nc','a')
nc.foo = np.array('bar','U')
nc.close()

The attribute is first written using nc_put_att_text, but then is rewritten using nc_put_att_string. On the other hand, this works

from netCDF4 import Dataset
import numpy as np
nc = Dataset('test_att_convert.nc','w')
nc.foo = np.array(1,np.int8)
nc.close()
nc = Dataset('test_att_convert.nc','a')
nc.foo = np.array(1,np.float64)
nc.close()

So the C lib is able to handle conversions of numeric types.

[jswhit]

It is not necessary to close and reopen the file to trigger the segfault, simply writing an attribute as NC_CHAR (with nc_put_att_text) and then re-writing it as NC_STRING (with nc_put_att_string) will do it.

[jswhit]

Simple C program to trigger the segfault:

#include <netcdf.h>
int main() {
   int dataset_id;
   const char *attstring[1] = {"bar"};
   nc_create("test.nc", NC_NETCDF4, &dataset_id);
   nc_put_att_text(dataset_id, NC_GLOBAL, "foo", 3, "bar");
   nc_put_att_string(dataset_id, NC_GLOBAL, "foo", 1, attstring);
   nc_close(dataset_id);
}

[WardF]

Hi Jeff,

Thanks for the bug report and the C program; I'm going to take a look at this now.

[WardF]

Looking at this, it appears the problem isn't in overwriting the attribute string itself, but rather something is happening when nc_close is called. Still debugging, but thought I'd check it out. It's slower going than usual, as OSX has replaced gdb with lldb, and there's new syntax to learn.

[WardF]

The issue has been fixed and the fix is in the current master development branch. Any bugs I've introduced have yet to reveal themselves, as all tests pass. The issue was a dangling data pointer when overwriting the existing attribute. Closing out the issue.