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 https://www.unidata.ucar.edu/jira/browse/NCF-XXX .
fsync: Changed default in autotools config file; fsync must now be explicitely enabled instead of explicitely disabled. NCF-239
Fixed netCDF-4 bug where odometer code for libdap2 mishandled stride > 1. Bug reported by Ansley Manke. NCF-249
Fixed netCDF-4 bug where netCDF would ignore the HDF5 reference type in
the file, instead of rejecting the file as it should. NCF-29
Fixed netCDF-4 bug with particular order of creation of dimensions,
coordinate variables, and subgroups resulting in two dimensions with the
same dimension ID. NCF-244
Fixed netCDF-4 bug with a multidimensional coordinate variable in a
subgroup getting the wrong dimension IDs for its dimensions. NCF-247
Fixed bug with incorrect fixed-size variable offsets in header getting
written when schema changed for files created by parallel-netcdf. Thanks
to Wei-keng Liao for developing and contributing the fix. NCF-234
Fixed bug in handling old servers that do not do proper Grid to
Structure conversions. NCF-232
Replaced the oc library with oc2.0
Fix bug with nc_get_var1_uint() not accepting unsigned ints larger
than 2**31. NCF-226
Fix to convert occurrences of '/' in DAP names to %2f. NCF-223
Fix bug in netCDF-4 with scalar non-coordinate variables with same name
as dimensions. NCF-222
Fix bug in which calling netCDF-4 functions in which behavior that
should not depend on order of calls sometimes produces the wrong
results. NCF-217
Merged in nccopy additions from Martin van Driel to support -g and -v
options for specifying which groups or variables are to be copied.
NCF-216
Merged in parallel-netcdf bugs fixes from Greg Sjaardema. NCF-214
Modify ncgen so that if the incoming file has a special attribute, then
it is used to establish the special property of the netcdf file, but the
attribute is not included as a real attribute in the file. NCF-213.
Added library version info to the user-agent string so that the server
logs will be more informative. NCF-210
Added work around for bad servers that sometimes sends DAP dataset with
duplicate field names. NCF-208
Fixed bug with strided access for NC_STRING type. NCF-206
Prevented adding an invalid _FillValue attribute to a variable (with
nonmatching type or multiple values), to avoid later error when any
record variable is extended. NCF-190
Fix bug in which some uses of vlen within compounds causes HDF5 errors.
NCF-155
Fixed ncdump bug in display of data values of variables that use
multiple unlimited dimensions. NCF-144
Fix bug in which interspersing def_var calls with put_var calls can
lead to corrupt metadata in a netCDF file with groups and inherited
dimensions. NCF-134
Building shared libraries works with DAP and netCDF4 functionality.
NCF-205
32-and-64-bit builds are working under MinGW on Windows. NCF-112
Config.h for Windows compiles are included in the build. NCF-98
NetCDF-4 dependency on NC_MAX_DIMS has been removed. NCF-71
64-bit DLL's are produced on Windows. NCF-65
DLL Packaging issues are resolved. NCF-54
The CMake build system (with related ctest and cdash systems for
testing) has been integrated into netCDF-C. This allows for Visual
Studio-based builds in addition to gcc-based builds. This requires at
least CMake version 2.8.8. This replaces/supplements the cross-compiled
set of Visual-Studio compatible netCDF libraries introduced in netCDF
4.2.1-rc1.
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. NCF-112 NCF-57
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.
[NCF-110][NCF-109][NCF-5]
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. [NCF-150]
Fixed bug using nccopy to compress with shuffling that doesn't compress
output variables unless they were already compressed in the input file.
[NCF-162]
Fixed bug in 64-bit offset files with large records, when last record
variable requires more than 2**32 bytes per record. [NCF-164]
Fix bug in which passing a NULL path to nc_open causes failure.
[NCF-173]
Fixed ncgen bugs in parsing and handling opaque data.
Fixed ncdump bug, not escaping characters special to CDL in enumeration
labels. [NCF-169]
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.
[NCF-171]
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. [NCF-172]
Fixed bug with ncdump -t interpreting unit attribute without base time
as a time unit. [NCF-175]
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. [NCF-168]
Added description of configure flags to installation guide.
Clarified documentation of arguments to ncopen() and nccreate() and
their default values.
Fixed doxygen installation guide source file to preserve line breaks in
code and scripts. [NCF-174]
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 netcdf.pc.in 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 Makefiles. [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. [NCF-120]
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. [NCF-142]
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]) [NCF-157]
Separate the Fortran and C++ libraries and release the C library and
ncdump/ncgen/nccopy without Fortran or C++. [NCF-24]
Documentation mostly migrated to Doxygen, from Texinfo. [NCF-26]
Properly convert vara start/count parameters to DAP [NCF-105][NCF-106]
Fixed major wasted space from previous default variable chunk sizes
algorithm. [NCF-81]
Fixed bug in nccopy, in which compression and chunking options were
ignored for netCDF-4 input files. [NCF-79]
Fixed bug in ncgen in which large variables (more than 218 elements)
duplicates the first 218 values into subsequent chunks of data
[NCF-154].
Applied Greg Sjaardema's nccopy bug fix, not compressing output
variables f they were not already using compression on the input file
when shuffle specified. [NCF-162]
Fixed problem when a URL is provided that contains only a host name.
[NCF-103]
Fixed behavior of ncgen flags so that -o => -lb and, in the absence of
any other markers, make the default be -k1 [NCF-158]
Created a text INSTALL file for netCDF-4.2 release. [NCF-161]
Fixed bug in ncgen for vlen arrays as fields of compound types where
datalists for those types was improperly interpreted NCF-145.
Improve use of chunk cache in nccopy utility, making it practical for
rechunking large files. [NCF-85]
Fixed nccopy bug copying a netCDF-4 file with a chunksize for an
unlimited dimension that is larger than the associated dimension size.
[NCF-139]
Fixed nccopy bug when rechunking a netCDF-4 file with a chunkspec option
that doesn't explicitly specify all dimensions. [NCF-140]
Fixed bug in netCDF-4 files with non-coordinate variable with the same
name as a dimension. [NCF-141]
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.
[NCF-143]
Fixed ncgen bug with netCDF-4 variable-length constants (H/T to Lynton
Appel). [NCF-145]
Incorporated Peter Cao's performance fixes using HDF5 link iterator for
any group with many variables or types. [NCF-148]
Incorporated Constantine Khroulev's bug fix for invalid usage of
MPI_Comm_f2c in nc_create_par. [NCF-135]
Fixed turning off fill values in HDF5 layers when NOFILL mode is set in
netCDF-4 API (thanks to Karen Schuchardt). [NCF-151]
Fixed bug with scalar coordinate variables in netCDF-4 files, causing
failure with --enable-extra-tests [NCF-149]
Cleaned up the definition and use of nulldup. [NCF-92][NCF-93][NCF-94]
Fixed various '#include' bugs. [NCF-91][NCF-96][NCF-127]
v2 API functions modified to properly call the external API instead of
directly calling the netcdf-3 functions. [NCF-100]
Fixed problem with 64-bit offset format where writing more than 2**31
records resulted in erroneous NC_EINVALCOORDS error. [NCF-101]
Restored original functionality of ncgen so that a call with no flags,
only does the syntax check. [NCF-104]
Corrected misc. test bugs [NCF-107]
Modified ncdump to properly output various new types (ubyte, ushort,
uint, int64, and uint64). [NCF-111]
Fixed incorrect link flag for szip in configure.ac [NCF-116]
ncdump -t now properly parses ISO "T" separator in date-time strings.
[NCF-16]
ncdump -t "human time" functionality now available for attributes and
bounds variables [NCF-70]
In ncdump, add -g option to support selection of groups for which data
is displayed. [NCF-11]
Now supports bluefire platform [NCF-52]
ncdump now properly displays values of attributes of type NC_USHORT as
signed shorts [NCF-82]
Rename some code files so that there are no duplicate filenames.
[NCF-99]
Demonstration of netCDF-4 Performance Improvement with KNMI Data
[NCF-113]
Dimension size in classic model netCDF-4 files now allows larger sizes
than allowed for 64-bit offset classic files. [NCF-117]
ncdump now reports correct error message when "-x" option specifying
NcML output is used on netCDF-4 enhanced model input. [NCF-129]
Fixed bug causing infinite loop in ncdump -c of netCDF-4 file with
subgroup with variables using inherited dimensions. [NCF-136]
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. [NCF-20]
Detect from configure script when szlib is needed [NCF-21]
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.
[NCF-22]
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.
[NCF-23]
Fix bug in configure that did not make the search for the xdr library
depend on --enable-dap. [NCF-41]
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. [NCF-42]
Fix ncgen bug, not properly handling unsigned longlong parsing. [NCF-43]
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. [NCF-47]
Define uint type for unsigned integer, if not otherwise available.
Symptom was compile error involving uint in putget.c. [NCF-49]
Fix username+password handling in the DAP client code. [NCF-50]
Add test for handling parallel I/O problem from f77 when user forgets to
turn on one of the two MPI flags. [NCF-60]
Resolved "make check" problems when ifort compiler. Some "make install"
problems remain when using MPI and shared libraries. [NCF-61]
Fix problem with f90_def_var not always handle deflate setting when
compiler was ifort. [NCF-67]
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. [NCF-68]
Improve documentation on handling client-side certificates [NCF-48]
Document that array arguments, except in varm functions, must point to
contiguous blocks of memory. [NCF-69]
Get netCDF-4 tests working for DLLs generated with mingw. [NCF-6]
Make changes necessary for upgrading to HDF5 1.8.7 [NCF-66]
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
--default-chunks-in-cache.
Better default chunks size. Now chunks are sized to fit inside the
DEFAULT_CHUNK_SIZE (settable at configure time with
--with-default-chunk-size= option.)
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.
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) builds.
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.
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 --disable-utilities.
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 name.
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
http://www.unidata.ucar.edu/software/netcdf/docs/faq.html\#shared\_intro
Fixed ncdump crash that happened when no arguments were used.
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.
Added support for the ifc Fortran-90 compiler creating files "netcdf.d"
and "typesizes.d" (instead of ".mod" files).
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 interface.
Increased advisory limits in C netcdf.h and Fortran netcdf.inc for
maximum number of dimensions, variables, and attributes.
Changed C++ declarations "friend NcFile" to "friend class NcFile" in
cxx/netcdfcpp.h to conform to standard.
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 interfaces.
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.
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() functions.
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. nccreate_mp() ncopen_mp() nc_set_base_pe()
nc_inq_base_pe()
Fixed makefile URL for Win32 systems in INSTALL file.
Made test for UNICOS system in the configure script case independent.
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 "AbsoftProFortran").
Corrected use of non-integral-constant-expression in specifying size of
temporary arrays in file "libsrc/ncx_cray.c".
Added Compaq Alpha Linux workstation example to INSTALL file.
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 intended.
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.
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 ncopen(),
nccreate(), 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), macros.make.in and
ncgen/Makefile to support NEC SUPER-UX 7.2.
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 errors.
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.
he 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 INTEGER1 and INTEGER2 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).
