Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

1497 lines (1100 sloc) 50.416 kB
This file contains a high-level description of this package's
evolution. Releases are in reverse chronological order (most recent
first). Recent releases include references to Jira issue identifiers
for more information, where '[NCF-XXX]' refers to .
------- --------
4.2.1 Released 2012-??-??
Added a Nspecific NC_MMAP mode flag to modify
behavior of NC_DISKLESS.
4.2.1-rc1 Released 2012-06-18
Ported static and shared libraries (DLL's) for both
32- and 64-bit Windows, including support for DAP
remote access, with netCDF-3 and netCDF-4/HDF5 support
enabled. The environment for this build is
MSYS/MinGW/MinGW64, but the resulting DLLs may be used
with Visual Studio.
Implemented diskless files for all netCDF formats. For
nc_create(), diskless operation performs all
operations in memory and then optionally persists the
results to a file on close. For nc_open(), but only
for netcdf classic files, diskless operation caches
the file in-memory, performs all operations on the
memory resident version and then writes all changes
back to the original file on close.
Added MMAP support. If diskless file support is
enabled, then it is possible to enable implementation
of diskless files using the operating system's MMAP
facility (if available). The enabling flag is
"--enable-mmap". This is most useful when using
nc_open() and when only parts of files, a single
variable say, need to be read.
Added configure flag for --disable-diskless.
Added nccopy command-line options to exploit diskless
files, resulting in large speedups for some
operations, for example converting unlimited dimension
to fixed size or rechunking files for faster access.
Upgraded doxygen and man-page documentation for ncdump
and nccopy utilities, including new -w option for
diskless nccopy, with an example.
Modified Makefile to allow for concurrent builds and
to support builds outside the source tree, e.g.
'mkdir build; cd build; SOURCE-DIR/configure' where
SOURCE-DIR is the top-level source directory.
Fixed some netCDF-4 bugs with handling strings in
non-netCDF-4 HDF5 files.
Fixed bug using nccopy to compress with shuffling that
doesn't compress output variables unless they were
already compressed in the input file.
Fixed bug in 64-bit offset files with large records,
when last record variable requires more than 2**32
bytes per record.
Fix bug in which passing a NULL path to nc_open causes failure.
Fixed ncgen bugs in parsing and handling opaque data.
Fixed ncdump bug, not escaping characters special to CDL
in enumeration labels.
Fixed bug reading netCDF int into a C longlong or
writing from longlong to external int on 32-bit
platforms with classic format files. The upper 32
bits of the longlong were not cleared on read or used
on write.
Resolved some erroneous returns of BADTYPE errors and
RANGE errors due to conflating C memory types with
external netCDF types when accessing classic or 64-bit
offset files.
Fixed bug with ncdump -t interpreting unit attribute
without base time as a time unit.
Changed port for testing remote access test server to
increase reliability of tests.
Modified ncio mechanism to support multiple ncio
packages, so that it is possible to have e.g. posixio
and memio operating at the same time.
Generation of documentation is disabled by default. Use
--enable-doxygen to generate.
Added description of configure flags to installation
Clarified documentation of arguments to nc__open() and
nc__create() and their default values.
Fixed doxygen installation guide source file to
preserve line breaks in code and scripts.
Cleaned up a bunch of lint issues (unused variables,
etc.) and some similar problems reported by clang
static analysis.
Updated and fixed pkg-config source file
to work with separated netCDF language-specific
packages. Also fixed nc-config to call nf-config,
ncxx-config, and ncxx4-config for for backward
compatibility with use of nc-config in current
[NCF-165] [NCF-179]
4.2 Released 2012-03-19 (Note: Jira entries include reference to '[NCF-XX]')
Completely rebuilt the DAP constraint handling.
This primarily affects users who specify a DAP constraint
as part of their URL.
Fixed cause of slow nccopy performance on file systems
with many records and large disk block size or many
record variables, by accessing data a record at a time
instead of a variable at a time.
Performance improvement to DAP code to support
fetching partial variables into the cache; especially
important when using nc_get_var() API.
A partial variable is one that has ranges attached
to the projection variables (e.g. x[1:10][20:21])
Separate the Fortran and C++ libraries and release
the C library and ncdump/ncgen/nccopy without Fortran or C++.
Documentation mostly migrated to Doxygen, from Texinfo.
Properly convert vara start/count parameters to DAP
Fixed major wasted space from previous default variable
chunk sizes algorithm.
Fixed bug in nccopy, in which compression and chunking
options were ignored for netCDF-4 input files.
Fixed bug in ncgen in which large variables
(more than 2**18 elements) duplicates the first
2**18 values into subsequent chunks of data
Applied Greg Sjaardema's nccopy bug fix, not
compressing output variables f they were not already
using compression on the input file when shuffle
Fixed problem when a URL is provided that contains only a
host name.
Fixed behavior of ncgen flags so that -o => -lb
and, in the absence of any other markers, make
the default be -k1
Created a text INSTALL file for netCDF-4.2 release.
Fixed bug in ncgen for vlen arrays as fields
of compound types where datalists for those
types was improperly interpreted
[NCF-145] (but see NCF-155).
Improve use of chunk cache in nccopy utility, making it
practical for rechunking large files.
Fixed nccopy bug copying a netCDF-4 file with a
chunksize for an unlimited dimension that is
larger than the associated dimension size.
Fixed nccopy bug when rechunking a netCDF-4 file with
a chunkspec option that doesn't explicitly specify all
Fixed bug in netCDF-4 files with non-coordinate
variable with the same name as a dimension.
Incorporated Wei Huang's fix for bug where netCDF-4
sometimes skips over too many values before adding
fill values to an in-memory buffer.
Fixed ncgen bug with netCDF-4 variable-length
constants (H/T to Lynton Appel).
Incorporated Peter Cao's performance fixes using HDF5
link iterator for any group with many variables or
Incorporated Constantine Khroulev's bug fix for
invalid usage of MPI_Comm_f2c in nc_create_par.
Fixed turning off fill values in HDF5 layers when NOFILL
mode is set in netCDF-4 API (thanks to Karen
Fixed bug with scalar coordinate variables in netCDF-4
files, causing failure with --enable-extra-tests
Cleaned up the definition and use of nulldup.
Fixed various '#include' bugs.
v2 API functions modified to properly call the external API
instead of directly calling the netcdf-3 functions.
Fixed problem with 64-bit offset format where writing more
than 2**31 records resulted in erroneous NC_EINVALCOORDS error.
Restored original functionality of ncgen so that a call with
no flags, only does the syntax check.
Corrected misc. test bugs
Modified ncdump to properly output various new types
(ubyte, ushort, uint, int64, and uint64).
Fixed incorrect link flag for szip in
ncdump -t now properly parses ISO "T" separator in
date-time strings.
ncdump -t "human time" functionality now available for
attributes and bounds variables
In ncdump, add -g option to support selection of groups
for which data is displayed.
Now supports bluefire platform
ncdump now properly displays values of attributes
of type NC_USHORT as signed shorts
Rename some code files so that there are no duplicate filenames.
Demonstration of netCDF-4 Performance Improvement with KNMI Data
Dimension size in classic model netCDF-4 files now allows
larger sizes than allowed for 64-bit offset classic files.
ncdump now reports correct error message when "-x" option
specifying NcML output is used on netCDF-4 enhanced model input.
Fixed bug causing infinite loop in ncdump -c of netCDF-4
file with subgroup with variables using inherited
4.1.3 2011-06-17
Replace use of --with-hdf5= and other such configure
options that violate conventions and causes build
problems. Set environment variables CPPFLAGS,
LDFLAGS, and LD_LIBRARY_PATH instead, before running
configure script.
Detect from configure script when szlib is needed
Fix bug that can silently zero out portions of a file
when writing data in nofill mode beyond the end of a
file, crossing disk-block boundaries with region to be
written while in-memory buffer is in a specific
state. This bug was observed disabling fill mode
using Lustre (or other large blksize file system) and
writing data slices in reverse order on disk.
Fix bug that prevents netCDF-4/HDF5 files created with
netCDF-4.1.2 from being read by earlier versions of
netCDF or HDF5 versions before 1.8.7.
Fix bug in configure that did not make the search for
the xdr library depend on --enable-dap.
Fix ncgen bug that did not use the value of a _Format
attribute in the input CDL file to determine the kind
of output file created, when not specified by the -k
command-line flag.
Fix ncgen bug, not properly handling unsigned longlong parsing.
Fix DAP client code to suppress variables with names
such as "x.y", which DAP protocol interprets as
variable "y" inside container "x". Such variables
will be invisible when accessed through DAP client.
Define uint type for unsigned integer, if not
otherwise available. Symptom was compile error
involving uint in putget.c.
Fix username+password handling in the DAP client code.
Add test for handling parallel I/O problem from f77
when user forgets to turn on one of the two MPI flags.
Resolved "make check" problems when ifort compiler.
Some "make install" problems remain when using MPI and
shared libraries.
Fix problem with f90_def_var not always handle deflate
setting when compiler was ifort.
Check that either MPIIO or MPIPOSIX flag is set when
parallel create or open is called. Also fix examples
that didn't set at least one of these flags.
Improve documentation on handling client-side certificates
Document that array arguments, except in varm
functions, must point to contiguous blocks of memory.
Get netCDF-4 tests working for DLLs generated with mingw.
Make changes necessary for upgrading to HDF5 1.8.7
4.1.3-rc1 2011-05-06
Stop looking for xdr if --disable-dap is used.
Don't try to run (some) fortran configure tests on
machines with no fortran.
Allow nccopy to rechunk with chunksizes larger than
current dimension lengths.
Initial implementation of CDMREMOTE is complete; needs
comprehensive testing.
4.1.3-beta1 2011-04-29
Fixed szlib not linking bug.
Fixed dreaded "nofill bug", lurking in netCDF classic
since at least 1999. Writing more than a disk
block's worth of data that crossed disk block
boundaries more than a disk block beyond the end of
file in nofill mode could zero out recently written
earlier data that hadn't yet been flushed to disk.
Changed setting for H5Pset_libver_bounds to ensure
that all netCDF-4 files can be read by HDF5 1.8.x.
Merged libncdap3 and libncdap4 into new libdap2 library.
The suffix dap2 now refers to the dap protocol. This is
in prep for adding dap4 protocol support.
Took out --with-hdf5 and related options due to high
cost of maintaining this non-standard way of finding
4.1.2 2011-03-29
Changes in build system to support building dlls on
Changes to fix portability problems and get things
running on all test platforms.
Some minor documentation fixes.
Fixed opendap performance bug for nc_get_vars; required
adding nc_get_var{s,m} to the dispatch table.
Now check for libz in
Fixed some bugs and some performance problems with
default chunksizes.
4.1.2-beta2 2011-01-11
Add "-c" option to nccopy to specify chunk sizes used
in output in terms of list of dimension names.
Rewrite netCDF-4 attribute put code for a large
speedup when writing lots of attributes.
Fix nc-config --libs when static dependent libraries
are not installed in the same directory as netCDF
libraries (thanks to Jeff Whitaker).
Build shared libraries by default, requiring separate
Fortran library. Static libraries now built only with
Refactor of HDF5 file metadata scan for large speedup
in opening files, especially large files.
Complete rewrite of the handling of character datalist
constants. The heuristics are documented in ncgen.1.
Eliminate use of NC_MAX_DIMS and NC_MAX_VARS in ncdump
and nccopy, allocating memory as needed and reducing
their memory footprint.
Add documentation for new nc_inq_path() function.
Use hashing to speedup lookups by name for files with
lots of dimensions and variables (thanks to Greg
Add options to nccopy to support uniform compression
of variables in output, shuffling, and fixing
unlimited dimensions. Documented in nccopy.1 man page
and User's Guide.
4.1.2-beta1 2010-07-09
Fix "ncdump -c" bug identifying coordinate variables in groups.
Fix bug in libsrc/posixio.c when providing sizehint
larger than default, which then doesn't get used
(thanks to Harald Anlauf).
Fix netCDF-4 bug caused when doing enddef/redef and
then defining coordinate variable out of order.
Fixed bug in man4 directory automake file which caused
documentation to be rebuilt after make clean.
Turned off HDF5 caching when parallel I/O is in use
because of its memory use.
Refactoring of netCDF code with dispatch layer to
decide whether to call netCDF classic, netCDF-4, or
opendap version of a function.
Refactoring of netCDF-4 memory internals to reduce
memory use and end dependence on NC_MAX_DIMS and
Modified constraint parser to be more compatible with
a java version of the parser.
Modified ncgen to utilize iterators internally; should be
no user visible effect.
Fixed two large-file bugs with using classic format or
64-bit offset format and accessing multidimensional
variables with more than 2**32 values.
4.1.1 2010-04-01
Fixed various build issues.
Fixed various memory bugs.
Fixed bug for netCDF-4 files with dimensions and coord
vars written in different orders, with data writes
Added test for HDF5-1.8.4 bug.
Added new C++ API from Lynton Appel.
4.1 2010-01-30
Much better memory leak checking with valgrind.
Added per-variable chunk cache control for better
performance. Use nc_set_var_chunk_cache /
nf_set_var_chunk_cache / nf90_set_var_chunk_cache to
set the per-variable cache.
Automatically set per-variable chunk cache when
opening a file, or creating a variable, so that the
cache is big enough for more than one chunk. (Can be
overridden by user). Settings may be changed with
configure options --max-default-chunk-size and
Better default chunks size. Now chunks are sized to
fit inside the DEFAULT_CHUNK_SIZE (settable at
configure time with --with-default-chunk-size=
Added nccopy utility for converting among netCDF
format variants or to copy data from DAP servers to
netCDF files.
The oc library has been modified to allow the occurrence
of alias definitions in the DAS, but they will be ignored.
The old ncgen has been moved to ncgen3 and
ncgen is now the new ncgen4.
Modified --enable-remote-tests to be on by default.
Fixed the nc_get_varm code as applied to DAP data sources.
Added tests for nc-config.
Many documentation fixes.
Added capability to use the parallel-netcdf
(a.k.a. pnetcdf) library to perform parallel I/O on
classic and 32-bit offset files. Use the NC_PNETCDF
mode flag to get parallel I/O for non-netcdf-4 files.
Added libcf library to netCDF distribution. Turn it on
with configure option --with-libcf.
Added capability to read HDF4 files created with the
SD (Scientific Data) API.
The DAP support was revised to closely mimic the
original libnc-dap support.
Significantly revised the data handling mechanism
in ncgen4 to more closely mimic the output from the
original ncgen.
Added prototype NcML output capability to ncgen4.
It is specified by the -lcml flag.
Added capability to read HDF5 files without dimension
scales. This will allow most existing HDF5 datasets to
be read by netCDF-4.
Fixed bug with endianness of default fill values for
integer types when variables are created with a
non-native endiannesss and use the default fill value.
Significant refactoring of HDF5 type handling to
improve performance and handle complicated nesting of
types in cross-platform cases.
Added UDUNITS2 to the distribution. Use --with-udunits
to build udunits along with netcdf.
Made changes suggested by HDF5 team to relax
creation-order requirement (for read-only cases) which
allows HDF5 1.6.x files to be retrofitted with
dimension scales, and be readable to netCDF-4.
Handle duplicate type names within different groups in
ncdump. Fix group path handling in absolute and
relative variable names for "-v" option.
Added nc-config shell script to help users build
netCDF programs without having to figure out all the
compiler options they will need.
Fixed ncdump -s bug with displaying special attributes
for classic and 64-bit offset files.
For writers, nc_sync() now calls fsync() to flush data
to disk sooner.
The nc_inq_type() function now works for primitive types.
4.0.1 2009-03-26
Added optional arguments to F90 API to
nf90_open/create, nf90_create_var, and
nf90_inquire_variable so that all netCDF-4 settings
may be accomplished with optional arguments, instead
of separate function calls.
Added control of HDF5 chunk cache to allow for user
performance tuning.
Added parallel example program in F90.
Changed default chunking to better handle very large
Made contiguous the default for fixed size data sets
with no filters.
Fixed bug in nc_inq_ncid; now it returns NC_ENOGRP if
the named group is not found.
Fixed man pages for C and F77 so that netCDF-4 builds
will result in man pages that document new netCDF-4
Added OPeNDAP support based on a new C-only implementation.
This is enabled using --enable-dap option and requires
libcurl. The configure script will attempt to locate libcurl,
but if it fails, then its location must be specified by the
--with-curl option.
4.0.1-beta2 2008-12-26
Changed chunksizes to size_t from int.
Fixed fill value problem from F77 API.
Fixed problems in netcdf-4 files with
multi-dimensional coordinate variables.
Fixed ncgen to properly handle CDL input that uses
Windows line endings ("\r\n"), instead of getting a
syntax error.
Added "-s" option to ncdump to display performance
characterisitics of netCDF-4 files as special virtual
attributes, such as _Chunking, _DeflateLevel, _Format,
and _Endianness.
Added "-t" option to ncdump to display times in human
readable form as strings. Added code to interpret
"calendar" attribute according to CF conventions, if
present, in displaying human-readable times.
Added experimental version of ncgen4 capable of
generating netcdf-4 data files and C code for creating
them. In addition, it supports the special attributes
_Format, etc.
4.0.1-beta1 2008-10-16
Fixed Fortran 90 int64 problems.
Rewrote HDF5 read/write code in accordance with
performance advice from Kent.
Fixed memory leaks in gets/puts of HDF5 data.
Fixed some broken tests for parallel I/O (i.e. MPI)
Fixed some cross-compile problems.
Rewrote code which placed bogus errors on the HDF5
error stack, trying to open non-existant attributes
and variables. Now no HDF5 errors are seen.
Removed man subdirectory. Now man4 subdirectory is
used for all builds.
Changed build so that users with access to parallel
make can use it.
Added experimental support for accessing data through
OPeNDAP servers using the DAP protocol (use
--enable-opendap to build it).
Fixed ncdump bugs with array field members of compound
type variables. Fixed ncdump bug of assuming default
fill value for data of type unsigned byte.
4.0 2008-05-31
Introduced the use of HDF5 as a storage layer, which
allows use of groups, user-defined types, multiple
unlimited dimensions, compression, data chunking,
parallel I/O, and other features. See the netCDF
Users Guide for more information.
3.6.3 2008-05-31
In ncdump and ncgen, added CDL support for UTF-8
encoding of characters in names and for escaped
special chars in names. Made sure UTF-8 names are
normalized using NFC rules before storing or
Handle IEEE NaNs and infinities in a
platform-independent way in ncdump output.
Added support for ARM representation of doubles,
(thanks to Warren Turkal).
Fixed bug in C++ API creating 64-bit offset
files. (See
Fixed bug for variables larger than 4 GB. (See
Changed the to build either 3.6.x or 4.x
build from the same
Build now checks gfortran version and handles it
cleanly, also Portland Group in Intel fortran, with
various configurations.
A Fortran file is now created at build time, based
on the setting of --disable-v2.
Documentation has been fixed in several places.
Upgraded to automake 1.10, autoconf 2.62, and libtool 2.2.2.
Includes missing Windows Visual Studio build files.
Fixed missing include of config.h in a C++ test program.
Fixed maintainer-clean in man directory.
Fixed --enable-c-only and make check.
Fixed behavior when opening a zero-length file.
Many portability enhancements to build cleanly on
various platforms.
Turned on some old test programs which were not being
used in the build.
3.6.2 2007-03-05
3.6.2 beta6 2007-01-20
Fine tuning of build system to properly handle cygwin,
Mingw, and strange configuration issues.
Automake 1.10 has a problem with running our tests on
MinGW, so I'm switching back to automake 1.9.6 for
this release.
3.6.2 beta5 2006-12-30
Now netCDF configuration uses autoconf 2.61, and
automake 1.10. (Thanks to Ralf Wildenhues for the
patches, and all the autotools help in general!)
Final major revision of netCDF tutorial before the
3.6.2 release.
Now netCDF builds under MinGW, producing a windows DLL
with the C and F77 APIs. Use the --enable-shared --enable-dll
--disable-cxx --disable-f90 flags to configure. (C++
and F90 have never been built as windows DLLs, but
might be in a future release if there is user
interest). This has all been documented in the netCDF
Porting and Installation Guide.
Now extreme numbers (i.e. those close to the limits of
their type) can be turned off in nc_test/nf_test, with
--disable-extreme-numbers. It is turned off
automatically for Solaris i386 systems.
Added --enable-c-only option to configure. This causes
only the core netCDF-3 C library to be built. It's the
same as --disable-f77 --disable-cxx --disable-v2
Added --disable-utilities to turn off building and
testing of ncgen/ncdump.
Fix a long-standing bug in nf90_get_att_text() pointed
out by Ryo Furue, to make sure resulting string is
blank-padded on return. This is fixed in the
Fortran-90 interface, but is impractical to fix in the
Fortran-77 interface implemented via cfortran.h.
Now large file tests are run if --enable-large-file-tests
is used in the configure.
For Cray users, the ffio module is used if the
--enable-ffio option is passed to configure.
Unrolled loops in byte-swapping code used on
little-endian platforms to reduce loop overhead. This
optimization resulted in a 22% speedup for some
applications accessing floats or ints (e.g. NCO
utilities ncap and ncbo) and a smaller speedup for
shorts or doubles.
Added "-k" option to ncdump and ncgen, for identifying
and specifying the kind of netCDF file, one of
"classic", "64-bit-offset", "hdf5", or "hdf5-nc3".
Removed output of kind of netCDF file in CDL comment
produced by ncdump.
Fixed bug of ncdump seg-faulting if invoked
incorrectly with option like "-c" or "-h" but no file
3.6.2 beta4 2006-08-15
Changed F77/F90 man pages from netcdf.3f and
netcdf.3f90 to netcdf_f77.3 and netcdf_f90.3. Also
fixed broken install of man pages.
Changed configure script so that "-g -O2" is no longer
set as CFLAGS, CXXFLAGS, and FFLAGS by default if a
GNU compiler is being used. Now nothing is set.
Changed configure script so that fortran flag is set
in config.h.
Updated Installation and Porting Guide, C++
Interface Guide, F77 and F90 Interface Guides.
Build with static libraries by default.
Added configure option --enable-separate-fortran,
which causes the fortran library to be built
separately. This is turned on automatically for shared
Improved clarity of error messages.
Changed configuration to get cygwin DLL and mingw DLL
builds working, for the C library only (i.e. no
F77, F90, or C++ APIs).
Changed type of ncbyte in C++ interface from unsigned
char to signed char, for consistency with C interface.
The C++ documentation warned this change would
eventually occur.
Changed the C++ interface to use only the netCDF-3 C
interface instead of the older netCDF-2 C interface.
This has the added benefit that on-the-fly numeric
conversions are now supported using get methods, for
example you can get data of any type as double. When
using --disable-v2 flag to configure, the C++
interface can now be built and installed.
3.6.2 beta3 2006-05-24
Changed to use default prefix of /usr/local instead of
package-based prefix of previous releases of
netCDF. Use the --prefix argument to the configure
script to override the default.
Made separate fortran library file, instead of
appending fortran library functions to the C library
file, if --enable-separate-fortran is used during
configure (it's turned on automatically if
--enable-shared is used). If uses, the fortran API
users must link to *both* the C library and the new
fortran library, like this: -lnetcdff -lnetcdf
Added netCDF examples in C, C++, F77, F90, and
CDL. See the examples subdirectory.
Added the NetCDF Tutorial.
Minor fixes to some of the netCDF documentation.
Made it possible to build without V2 API using
--disable-v2 from configure.
Switched to new build system, with automake and
libtool. Now shared libraries are built (as well as
static ones) on platforms which support it. For more
information about shared libraries, see
Fixed ncdump crash that happened when no arguments were
Fixed for building with gfortran 4.1.0.
Important fix for machines whose SIZEOF_SIZE_T !=
SIZEOF_LONG, such as NEC-SX, thanks to Stephen Leak.
Fixed C++ on AIX platform.
Fixed 64-bit builds on AIX platform.
Removed bad assertion that could be triggered in rare
cases when reading a small file.
Added comments in v1hpg.c to clarify purpose of each
internal function.
Make sure filesize is determined in nc_close() *after*
buffers get flushed.
Fix long-standing problem resulting in files up to 3
bytes longer than necessary if there is exactly one
record variable of type byte, char, or short and if
the number of values per record for that variable is
not divisible by 4 (or 2 in the case of short). Now
the filesize determined from header info by
NC_calcsize should be correct in all cases.
3.6.1 2006-01-31
Updated installation manual for 3.6.1.
Changed installation to try to provide correct
compiler flags for compiling in 64-bit mode on Sun,
Irix, AIX, and HPUX. (HPUX doesn't work for me,
however). Now run configure with --enable-64bit to get
a 64 bit compile.
Fixed long-standing bug that would cause small netCDF
files to be padded on the end with zero bytes to 4096
bytes when they were opened and changed. Now small
files should stay small after you change a value.
Fixed bug in assertions in putget.c that would only be
noticed if you change the manifest constant
NC_MAX_DIMS in netcdf.h to be different from
Moved test ftest.F from fortran to nf_test directory,
and fixed bug in ftest.F which caused it to return 0
even if tests failed (no tests were failing,
however). Also renamed some test
output files to make things a little clearer.
If open for writing, pad with up to 3 extra zero bytes
before close to the correct canonical length,
calculated from the header. Previously files could be
short due to not padding when writing in NOFILL mode.
Doubled arbitrary limits on number of dimensions,
variables, attributes, and length of names.
Change name of nc_get_format() to nc_inq_format().
Add analogous interfaces for nf_inq_format(),
nf90_inquire(), and NcFile::get_format() to f77, f90,
and C++ interfaces. Document new function in texinfo
files. Add minimal test to nc_test, nf_test.
3.6.1-beta3 2005-02-17
Added function nc_get_format(int ncid, int* formatp)
that returns either NC_FORMAT_CLASSIC or
NC_FORMAT_64BIT for a CDF1 or CDF2 file, respectively.
Added test to nc_test that detects whether format
version was changed after a file is reopened and
define mode is entered.
Correctly configure for Intel ifort Fortran compiler on Linux.
3.6.0-p1 2005-02-18
Fixed bug that changes CDF2 files to CDF1 files if CDF2
file is reopened for write access and either an
attribute is changed or define mode is entered.
3.6.1-beta2 2005-1-6
Fixed absoft compile problem. Maybe.
3.6.1-beta1 2005-1-3
Fixed Cygwin C++ problem.
Fixed large file problem in MS Visual C++.NET environment.
More information in installation and porting guide.
3.6.0 2004-12-16
Added texinfo source for the documentation.
Added large file tests to Windows directory in distribution.
Modified win32 visual studio project files so that m4
is no longer required to build netcdf under visual studio.
Modified rules.make to use install instead of cp,
fixing install problem for cygwin users.
Modified configure/install stuff to support HP-UX.
Modified configure/install stuff to support G95.
In the f90 interface, applied Arnaud Desitter's fixes
to correct mismatches between scalar and array
arguments, eliminating (legitimate) complaints by the
NAGWare f95 compiler. Also fixed bugs introduced in
3.6.0-beta5 in the mapped array interfaces.
3.6.0-beta6 2004-10-05
Fixed AIX 64-bit/largefile install problems.
Removed FAQ section from netcdf.texi User's Guide, in
deference to online version that can be kept up to
date more easily.
3.6.0-beta5 2004-10-04
Fixed assertion violation on 64-bit platforms when
size of last fixed size variable exceeds 2^32 - 1.
Removed another restriction on file size by making
record size (derived from other sizes, not part of the
format) an off_t instead of a size_t, when an off_t is
larger than a size_t. This permits records to be
*much* larger in either classic format or
64-bit-offset format.
Incorporated patch from Mathis Rosenhauer to improve
performance of Fortran 90 interface for calls to
nf90_put_var_TYPE(), nf90_get_var_TYPE(),
nf90_put_vara_TYPE(), and nf90_get_vara_TYPE()
functions by not emulating them with the corresponding
nf90_put_varm_TYPE() and nf90_get_varm_TYPE() calls.
Added tests for invalid offsets in classic format when
defining multiple large variables.
Improved installation ease. Have configure script use
Large File Support as a default, if available.
Add "extra_test" as a target for testing Large File
3.6.0-beta3 2004-08-24
Upgraded to recent autoconf, changed configure to
(hopefully) improve installation. Also added macros
to deal with large file systems.
Added nf_set_default_format to Fortran interface.
Added testing to the set_default_format functions to
nc_test and nf_test.
Added documentation to the man page for
set_default_format functions.
Added two new error return codes to C, f77, and f90
interfaces for invalid dimension size and for bad
variable size. Made test for max dimension size
depend on whether 64-bit offsets used. Fixed bug with
dimension sizes between 2^31 and 2^32 (for byte
Fixed ncdump to properly print dimensions larger than
Fixed ncgen to properly handle dimensions between 2^31
and 2^32.
Added -v2 (version 2 format with 64-bit offsets)
option to ncgen, to specify that generated files or
generated C/Fortran code should create 64-bit offset
files. Also added -x option to ncgen to specify use
of no-fill mode for fast creation of large files.
Added function to set default create mode to C
interface (nc_set_default_create).
Added win32 directory, with NET subdirectory to hold
.NET port of netCDF. To use, open netcdf.sln with
Visual Studio, and do a clean and then a build of
either the debug or release builds. Tests will be run
as part of the build process. VC++ with managed
extensions is required (i.e. VC++.NET).
Added windows installer files to build windows binary
3.6.0-beta1 By incorporating Greg Sjaardema's patch, added support
for 64-bit offset files, which remove many of the
restrictions relating to very large files (i.e.
larger than 2 GB.) This introduces a new data format
for the first time since the original netCDF format
was introduced. Files in this new 64-bit offset
format can't be read by earlier versions of
netCDF. Users should continue to use the netCDF
classic format unless they need to create very large
The test suite, nc_test, will now be run twice, once for
netCDF classic format testing, and once for 64-bit offset
format testing.
The implementation of the Fortran-77 interface has been
adapted to version 4.3 of Burkhard Burow's "cfortran.h".
Added NEC SX specific optimization for NFILL tunable
parameter in libsrc/putget.c
Added support for the ifc Fortran-90 compiler creating
files "netcdf.d" and "typesizes.d" (instead of ".mod"
Fixed access to iargc and getarg functions from
Fortran-90 for NAG f90 compiler, contributed by Harald
3.5.1 2004-02-03
Updated INSTALL.html for Mac OS X (Darwin).
Made the installation of the netCDF Fortran-90 module
file more robust regarding the name of the file.
Added support for eight-byte integers in Fortran90
Increased advisory limits in C netcdf.h and Fortran for maximum number of dimensions,
variables, and attributes.
Changed C++ declarations "friend NcFile" to "friend
class NcFile" in cxx/netcdfcpp.h to conform to
Added Dan Schmitt's backward compatible extension to
the C++ record interface to work with arbitrary
dimension slices.
Added C++ documentation note that caller is
responsible for deleting pointer returned by
Variable::values() method when no longer needed.
Made C++ interface more standard; the result may not
compile on some old pre-standard C++ compilers.
Fixed bug in ncgen when parsing values of a
multidimensional char variable that resulted in
failure to pad a value with nulls on IRIX.
Fixed ncdump bug adding extra quote to char variable
data when using -fc or -ff option.
Fixed so compiling with -DNO_NETCDF_2 will work for
building without backward-compatibility netCDF-2
Eliminated use of ftruncate(), because it fails on
FAT32 file systems under Linux.
Initialized a pointer in putget.m4 (used to generate
putget.c) that was involved in uninitialized memory
references when nc_test is run under Purify. Two
users had reported seeing crashes resulting from this
problem in their applications.
Reverted pointer initializations in putget.m4, after
testing revealed these caused a performance problem,
resulting in many extra calls to px_pgin and px_pgout
when running nc_test.
Added checking of size of "dimids" vector in function
nf90_inquire_variable(...) and error-returning if it
isn't sufficiently capacious.
Added variable index to ncvarget() and ncattinq() error
messages and attribute name to ncattinq() error message.
Tweaked configure script to work with recent C++ compilers.
Fixed a memory leak in C++ interface, making sure
NcVar::cur_rec[] gets deleted in NcVar destructor.
Reimplemented nc_sync() fix of version 3.5.0 to eliminate
performance penalty when synchronization is unnecessary.
Changed order of targets in Makefile to build Fortran
interface last, as a workaround for problem with make
on AIX platforms.
3.5.0 2001-03-23
Added Fortran 90 interface.
Changed C macro TIMELEN in file cxx/nctst.cpp to
TIMESTRINGLEN to avoid clash with macro defined on AIX
systems in /usr/include/time.h.
Fixed miswriting of netCDF header when exiting define
mode. Because the header was always written correctly
later, this was only a problem if there was another
reader of the netCDF file.
Fixed explicit synchronizing between netCDF writer and
readers via the nc_sync(), nf_sync(), and ncsync()
Fixed a number of bugs related to attempts to support
shrinking the header in netCDF files when attributes
are rewritten or deleted. Also fixed the problem that
nc__endef() did not work as intended in reserving
extra space in the file header, since the extra space
would be compacted again on calling nc_close().
Fixed the "redef bug" that occurred when nc_enddef()
or nf_enddef() is called after nc_redef() or
nf_redef(), the file is growing such that the new
beginning of a record variable is in the next "chunk",
and the size of at least one record variable exceeds
the chunk size (see netcdf.3 man page for a
description of this tuning parameter and how to set
it). This bug resulted in corruption of some values
in other variables than the one being added.
The "__" tuning functions for the Fortran interface,
nf__create, nf__open, and nf__enddef, are now
documented in the Fortran interface man pages.
Add an 'uninstall' target to all the Makefiles.
Dave Glowacki <dglo@SSEC.WISC.EDU> 199810011851.MAA27335
Added support for multiprocessing on Cray T3E.
Hooks added by Glenn, but the majority of the work
was done at NERSC. Also includes changes to ffio
option specification. Patch rollup provided by
R. K. Owen <rkowen@Nersc.GOV>. The following functions
are added to the public interface.
Fixed makefile URL for Win32 systems in INSTALL file.
Made test for UNICOS system in the configure script case
Ported to the following systems:
AIX 4.3 (both /bin/xlc and /usr/vac/bin/xlc compilers)
IRIX 6.5
IRIX64 6.5
Changed the extension of C++ files from ".cc" to ".cpp".
Renamed the C++ interface header file "netcdfcpp.h"
instead of "netcdf.hh", changing "netcdf.hh" to
include "netcdfcpp.h" for backward compatibility.
Treat "FreeBSD" systems the same as "BSD/OS" system
w.r.t. Fortran and "whatis" database.
Corrected manual pages: corrected spelling of "enddef" (was
"endef") and ensured that the words "index" and "format"
will be correctly printed.
Updated support for Fortran-calling-C interface by
updating "fortran/cfortran.h" from version 3.9 to
version 4.1. This new version supports the Portland
Group Fortran compiler (C macro "pgiFortran")
and the Absoft Pro Fortran compiler (C macro
Corrected use of non-integral-constant-expression
in specifying size of temporary arrays in file
Added Compaq Alpha Linux workstation example to INSTALL
Ported cfortran.h to Cygnus GNU Win32 C compiler (gcc
for Windows).
Fixed bug in ncdump using same CDL header name when
called with multiple files.
Added new NULL data type NC_NAT (Not A Type) to
facilitate checking whether a variable object has had
its type defined yet, for example when working with
packed values.
Fixed use of compile-time macro NO_NETCDF_2 so it
really doesn't include old netCDF-2 interfaces, as
Ported to MacOS X Public Beta (Darwin 1.2/PowerPC).
Fixed C++ friend declarations to conform to C++ standard.
Changed INSTALL file to INSTALL.html instead.
3.4 1998-03-09
Fixed ncx_cray.c to work on all CRAY systems,
not just CRAY1. Reworked USE_IEG, which was incorrect.
Reworked short support. Now USE_IEG and otherwise
both pass t_ncx.
To better support parallel systems, static and malloc'ed
scratch areas which were shared in the library
were eliminated. These were made private and on the
stack where possible. To support this, the macros
ALLOC_ONSTACK and FREE_ONSTACK are defined in onstack.h.
The buffered i/o system implementation in posixio.c
was reimplemented to limit the number and size of read()
or write() system calls and use greater reliance on memory
to memory copy. This saves a great deal of wall clock time
on slow (NFS) filesystems, especially during nc_endef().
Added performance tuning "underbar underbar" interfaces
nc__open(), nc__create(), and nc__enddef().
The 'sizehint' contract between the higher
layers and the ncio layer is consistently enforced.
The C++ interface has been updated so that the
deprecated "nclong" typedef should no longer be
required, and casts to nclong no longer necessary. Just
use int or long as appropriate. nclong is still
supported for backwards compatibility.
The ncdump utility now displays byte values as signed,
even on platforms where the type corresponding to a C
char is unsigned (SGI, for example). Also the ncdump
and ncgen utilities have been updated to display and
accept byte attributes as signed numeric values (with a
"b" suffix) instead of using character constants.
In libsrc/error.c:nc_strerror(int), explain that
NC_EBADTYPE applies to "_FillValue type mismatch".
Some changes to configure scripts (aclocal.m4), and ncgen/Makefile to support
The "usage" messages of ncgen and ncdump include the
string returned from nc_inq_libvers().
Corrected some casts in the library so that all phases of
the arithmetic computing file offsets occurs with "off_t"
type. This allows certain larger netcdf files to be created
and read on systems with larger (64bit) off_t.
In ncgen, multidimensional character variables are now
padded to the length of last dimension, instead of just
concatenating them. This restores an undocumented but
convenient feature of ncgen under netCDF-2. Also, a
syntax error is now reliably reported if the netcdf name
is omitted in CDL input.
Fortran and C code generated by ncgen for netCDF
components whose names contain "-" characters will now
compile and run correctly instead of causing syntax
The library allows "." characters in names as well
as "_" and "-" characters. A zero length name "" is
explicitly not allowed. The ncgen utility will now
permit "." characters in CDL names as well.
Memory leaks in the C++ interface NcVar::as_*() member
functions and NcFile::add_var() member function are
fixed. The documentation was fixed where it indicated
incorrectly that the library managed value blocks that
the user is actually responsible for deleting.
The values of the version 2 Fortran error codes have
been modified to make the version 2 Fortran interface
more backward compatible at the source level.
Added support for systems whose Fortran INTEGER*1 and
INTEGER*2 types are equivalent to the C "long" type but
whose C "int" and "long" types differ. An example of
such a system is the NEC SX-4 with the "-ew" option to
the f90 compiler (sheesh, what a system!).
Fixed Version 2 Fortran compatibility bug: NCVGTG,
NCVGGC, NCVPTG, and NCVPGC didn't work according to the
Version 2 documentation if the innermost mapping value
(i.e. IMAP[1]) was zero (indicating that the netCDF
structure of the variable should be used).
3.3.1 1997-06-16
One can now inquire about the number of attributes that a
variable has using the global variable ID.
The FORTRAN interface should now work on more systems.
In particular:
It should now work with FORTRAN compilers whose
"integer*1" datatype is either a C "signed char",
"short", or "int" and whose "integer*2" datatype is
either a C "short" or "int".
It should now work with FORTRAN compilers that are
extremely picky about source code formatting (e.g.
the NAG f90 compiler).
The dependency on the non-POSIX utility m4(1) for
generating the C and FORTRAN manual pages has been
EXTERNAL statements have been added to the FORTRAN
include-file "" to eliminate excessive
warnings about "unused" variables (which were actually
functions) by some compilers (e.g. SunOS 4.1.3's f77(1)
version 1.x).
Building the netCDF-3 package no longer requires the
existence of the Standard C macro RAND_MAX.
Fixed an ncdump bug resulting in ncdump reporting
Attempt to convert between text & numbers
when _FillValue attribute of a character variable set to
the empty string "".
Made ncgen tests more stringent and fixed various bugs
this uncovered. These included bugs in handling byte
attributes on platforms on which char is unsigned,
initializing scalar character variables in generated C
code under "-c" option, interspersing DATA statements
with declaration statements in generated Fortran code
under "-f" option, handling empty string as a value
correctly in generated C and Fortran, and handling
escape characters in strings. The Fortran output under
the "-f" option was also made less obscure and more
portable, using automatic conversion with netCDF-3
interfaces instead of "BYTE", "INTEGER*1", or
"INTEGER*2" declarations.
Fixed a C++ interface problem that prevented compiling
the C++ library with Digital's cxx compiler.
Made ncgen "make test" report failure and stop if test
resulted in a failure of generated C or Fortran code.
The file that you are now reading was created to contain
a high-level description of the evolution of the
netCDF-3 package.
3.3 1997-05-15
The production version of the netCDF-3 package was
A comparison of the netCDF-2 and netCDF-3 releases can
be found in the file COMPATIBILITY.
Jump to Line
Something went wrong with that request. Please try again.