- update to version 1.3.1:
* add parallel IO capabilities. netcdf-c and hdf5 must be compiled with MPI
support, and mpi4py must be installed. To open a file for parallel access,
use `parallel=True` in `Dataset.__init__` and optionally pass the mpi4py Comm instance
using the `comm` kwarg and the mpi4py Info instance using the `info` kwarg.
IO can be toggled between collective and independent using `Variable.set_collective`.
See `examples/mpi_example.py`. Issue #717, pull request #716.
Minimum cython dependency bumped from 0.19 to 0.21.
* Add optional `MFTime` calendar overload to use across all files, for example,
`'standard'` or `'gregorian'`. If `None` (the default), check that the calendar
attribute is present on each variable and values are unique across files raising
a `ValueError` otherwise.
* Allow _FillValue to be set for vlen string variables (issue #730).
- update to version 1.3.0:
* always search for HDF5 headers when building, even when nc-config is used
(since nc-config does not always include the path to the HDF5 headers).
Also use H5get_libversion to obtain HDF5 version info instead of
H5public.h. Fixes issue #677.
* encoding kwarg added to Dataset.__init__ and Dataset.filepath (default
is to use sys.getfilesystemencoding()) so that oddball
encodings (such as cp1252 on windows) can be handled in Dataset
filepaths (issue #686).
* Calls to nc_get_vars are avoided, since nc_get_vars is very slow (issue
#680). Strided slices are now converted to multiple calls to
nc_get_vara. This speeds up strided slice reads by a factor of 10-100
(especially for NETCDF4/HDF5 files) in most cases. In some cases, strided reads
using nc_get_vars are faster (e.g. strided reads over many dimensions
such as var[:,::2,::2,::2])), so a variable method use_nc_get_vars was added.
var.use_nc_get_vars(True) will tell the library to use nc_get_vars instead
of multiple calls to nc_get_vara, which was the default behaviour previous
to this change.
* fix utc offset time zone conversion in netcdftime - it was being done
exactly backwards (issue #685 - thanks to @pgamez and @mdecker).
* Fix error message for illegal ellipsis slicing, add test (issue #701).
* Improve timezone format parsing in netcdftime
* make sure numpy datatypes used to define CompoundTypes have
isalignedstruct flag set to True (issue #705), otherwise.
segfaults can occur. Fix required raising them minimum numpy requirement
from 1.7.0 to 1.9.0.
* ignore missing_value, _FillValue, valid_range, valid_min and valid_max
when creating masked arrays if attribute cannot be safely
cast to variable data type (and issue a warning). When setting
these attributes don't cast to variable dtype unless it can
be done safely and issue a warning. Issue #707.
- update to version 1.2.9:
* Fix for auto scaling and masking when _Unsigned attribute set (create
view as unsigned type after scaling and masking). Issue #671.
* Always mask values outside valid_min, valid_max (not just when
missing_value attribue present). Issue #672.
- remove check boundary condition
- Implement single-spec version
- Fix source URL.
- Update to version 1.2.8
* recognize _Unsigned attribute used by netcdf-java to designate unsigned
integer data stored with a signed integer type in netcdf-3 (issue #656).
* add Dataset init memory parameter to allow loading a file from memory
(pull request #652, issues #406 and #295).
* fix for negative times in num2date (issue #659).
* fix for failing tests in numpy 1.13 due to changes in numpy.ma (issue #662).
* Checking for _Encoding attribute for NC_STRING variables, otherwise use
'utf-8'. 'utf-8' is used everywhere else, 'default_encoding' global module
variable is no longer used. getncattr method now takes optional kwarg
'encoding' (default 'utf-8') so encoding of attributes can be specified
if desired. If _Encoding is specified for an NC_CHAR ('S1') variable,
the chartostring utility function is used to convert the array of
characters to an array of strings with one less dimension (the last
dimension is interpreted as the length of each string) when reading the
data. When writing the data, stringtochar is used to convert a numpy
array of fixed length strings to an array of characters with one more
dimension. chartostring and stringtochar now also have an 'encoding' kwarg.
Automatic conversion to/from character to string arrays can be turned off
via a new set_auto_chartostring Dataset and Variable method (default
is True). Addresses issue #654.
* Cython >= 0.19 now required, _netCDF4.c and _netcdftime.c removed from
repository.
- Update to version 1.2.7
* fix for issue #624 (error in conversion to masked array when variable slice
returns a scalar). This is a regression introduced in 1.2.5 associated
with support for vector missing_values. Test (tst_masked5.py) added for
vector missing_values.
* fix for python 3.6 compatibility (error retrieving character _FillValue attribute,
issue #626). Test with python 3.6 using travis CI.
- Update to version 1.2.6
* fix some test failures on big endian PPC64 that were due to
errors in byte-swapping logic. Also fixed bug in enum
code exposed on PPC64 (issue #608).
* remove support for python 2.6 (it probably still will work for a while
though).
* Sometimes checking that data being assigned to a variable has
an 'ndim' attribute is not sufficient, instead check to see that
the object supports the buffer interface (issue #613).
* make get_variables_by_attributes work in MFDataset (issue #610)
The hack is also applied for set_auto_maskandscale, set_auto_scale,
set_automask, so these don't have to be duplicated in MFDataset (pull
request #571).
- Update to version 1.2.5
* Add MFDataset.set_auto_maskandscale (plus set_auto_scale, set_auto_mask).
Fixes issue #570.
* Use valid_min/valid_max/valid_range attributes when defining mask
(issue #576). Values outside the valid range are considered to
be missing when defining the mask.
* Fix for issue #584 (add support for dates before -4712-1-1 in 360_day
and 365_day calendars to netcdftime.utime).
* Fix for issue #593: add support for datetime.timedelta operations
(adding and subtracting timedelta, subtracting two datetime
instances to compute time duration between them), implement
datetime.replace() and datetime.__str__(). datetime.__repr__()
includes the full state of an instance. Add datetime.calendar.
datetime comparison operators have full accuracy now.
* Fix for issue #585 by increasing the size of the buffer used to store the
filepath.
* Fix for issue #592: Add support for string array attributes. (When
reading, a vlen string array attribute is returned as a list of
strings. To write, use var.setncattr_string("name", ["two", "strings"]).)
* Fix for issue #596 - julian day calculations wrong for negative years,
caused incorrect rountrip num2date(date2num(date)) roundtrip for dates with year
< 0.
* Make sure negative years work in utime.num2date (issue #596).
* raise NotImplementedError when trying to pickle Dataset, Variable,
CompoundType, VLType, EnumType and MFDataset (issue #602).
* Fix for issue #527: initialize vldata[i].p in Variable._get(...).
- Update to version 1.2.4
* Fix for issue #554. It is now ensured that data is in native endian
byte order before passing to netcdf-c library. Data read from variable
with non-native byte order is also byte-swapped, so that dtype remains
consistent with netcdf variable. Behavior now consistent with h5py.
* raise warning for HDF5 1.10.x (issue #549), since backwards
incompatible files may be created.
* raise AttributeError instead of RuntimeError when attribute operation
fails. raise IOError instead of RuntimeError when nc_create or
nc_open fails (issue #546).
* Use NamedTemporaryFile instead of deprecated mktemp in tests
(pull request #543).
* add AppVeyor automated windows tests (pull request #540).
- Update to version 1.2.3.1
* fix bug in setup.py (pull request #539, introduced in issue #518).
- Update to version 1.2.3
* try to avoid writing NC_STRING attributes if possible, by
trying to convert unicode strings to ascii and write as NC_CHAR (issue
#529). This preserves compatibility with clients (like Matlab) that
can't deal with NC_STRING attributes. A 'setncattr_string' method
was added for Dataset and Variable to that users can force attributes
to be written as NC_STRING if necessary.
* fix failing tests with numpy 1.11 (issues #521 and #522).
* fix indentation bug in nc4tonc3 utility (issue #519).
* add the capability in setup.py to use pkg-config instead of
nc-config (pull request #518).
* make sure slices which return scalar masked arrays
are consistent with numpy.ma (issue #515).
* add test/tst_cdf5.py and test/tst_filepath.py (to test new
NETCDF3_64BIT_DATA format and filepath Dataset method).
* expose netcdftime.__version__ (issue #504).
* fix potential memory leak in Dataset.filepath in attempt to fix
mysterious segfaults on CentOS6 (issue #506). Segfaults
can apparently still occur on systems like CentOS6 with old versions of glibc.
- Update to version 1.2.2
* fix failing tests on python 2.6 (issue #497). Change minimum required
python from 2.5 to 2.6.
* Potential memory leaks fixed by freeing string pointers internally allocated
in netcdf-c using nc_free_string. Also use nc_free_vlens to free space allocated
for vlens inside netcdf-c (issue #495).
* invoke str on filename argument to Dataset constructor, so pathlib
instances can be used (issue #489).
* don't use hardwired NC_MAX_DIMS or NC_MAX_VARS internally to allocate space
for dimension or variable ids. Instead, find out the number of dims
and vars and use malloc. NC_MAX_NAME is still used to allocate space
for attribute and variable names, since there is no obvious way to
determine the length of these names.
* if trying to write a unicode attribute, check to see if it exists
first and is NC_CHAR, and if so, delete it and recreate it. Workaround for C
lib bug discovered in issue #485.
* support for NETCDF3_64BIT_DATA format supported in netcdf-c 4.4.0.
Similar to NETCDF3_64BIT (now NETCDF3_64BIT_OFFSET), but includes
64 bit dimensions and sizes, plus unsigned and 64 bit integer
data types.
* make sure chunksize does not exceed dimension size
(for non-unlimited dimensions) on variable creation (issue #480).
* add 'size' attribute to Dimension (same as len(d), where
d is a Dimension instance, issue #477).
* fix bug in nc3tonc4 with --unpackshort=1 (issue #474).
* dates do not have to be contiguous, i.e. can be before and after the
missing dates in Gregorian calendar (pull request #476).
- Update to version 1.2.1
* add the capability to slice variables with unsorted integer sequences,
or integer sequences with duplicates (issue #467). This was done
by converting boolean array slices to integer array slices internally,
instead of the other way around.
* raise TypeError if masked array assigned to a VLEN str variable slice
(issue #464).
* Ellipsis now can be used with scalar VLEN str variables (issue #458).
Slicing of scalar VLEN (non-str) variables now works.
* Allow non-positive reference years in non-real-world calendars
(issue #442).
- Update to version 1.2.0
* Fixes to setup.py for building on windows (issue #460).
* warnings now issued if file being read contains unsupported
variables or data types (they were previously being silently
skipped).
* added 'get_variables_by_attributes' method (issue #454).
* check for 'units' attribute in date2index (issue #453).
* added support for enum types (issue #452).
* added 'isopen' Dataset method (issue #450).
* raise ValueError if year 0 or negative year used in time units string.
The year 0 does not exist in the Julian and Gregorian
calendars (issue #442).
- Initial version