17-03-2012 Shahram Najm
-- released version 1.9.14
* Disabled boustrophedonic ordering in second order packing.
* Fixed a problem producing unwanted padding between messages when converting some fields from second order packing
* Improved thread safety in relation to the creation of an index. Creating a new grib_handle or indexing different files is thread safe.
* Updated the list of parameters
* Fixed minor bugs
17-03-2011 Enrico Fucile
-- released version 1.9.9
* Several changes have been made to the internal functions used by the interpolation package in EMOSLIB 381
to make the results of the grib_api based interpolation as close as possible to the GRIBEX based one.
* laplacianOperator for spectral fields is always computed. If the computation is not required please
set the key computeLaplacianOperator=0. Default behaviour is computeLaplacianOperator=1.
* New parameters have been added to the parameter database and to the definition files regarding
the MACC project.
* Second order packing (packingType=grid_second_order) is producing smaller fields as the default setting
is to use the boustrophedonic option.
25-02-2011 Enrico Fucile
-- released version 1.9.8
* WARNING: A significant change in the fortran interface has been done for which some
software using the previous versions could exhibits memory allocation problems due
to unreleased memory associated with the GRIB messages. In the previous versions
(starting from version 1.6.1) the argument IGRIB of the functions GRIB_NEW_FROM_FILE,
GRIB_NEW_FROM_MESSAGE, GRIB_NEW_FROM_INDEX, GRIB_CLONE was an input and output argument.
In the new version IGRIB is output only and the grib message associated with IGRIB will
never be released by the GRIB_NEW_FROM_FILE function. This means that to release the memory
associated to the message a call to GRIB_RELEASE(IGRIB) has to be introduced when the message
is not needed any more. In the previous version the memory was released when a value of IGRIB
associated with a valid message was passed to the function. It has been realised that this
behaviour was source of unexpected results when the IGRIB variables used in the user code were
not initialised to a non valid IGRIB value (zero or negative integer). We strongly advise the
users to revise the existing code using previous grib_api versions or to link to the new library,
add the appropriate GRIB_RELEASE calls and test it for memory allocation.
For convenience we report an example of code which could cause problems and has to be reviewed.
integer :: IGRIB1
integer :: IGRIB2
call grib_new_from_file(IGRIB1,...)
... do something ...
call grib_new_from_file(IGRIB2,...)
... do something ...
IGRIB1 and IGRIB2 are not initialised to zero and we don't know which value is assigned to them.
It could be IGRIB1=IGRIB2=5 or any other number greater or equal to zero. If this happens at the
second call of grib_new_from_file IGRIB1 will be replaced by IGRIB2 producing very dangerous
consequences as the IGRIB1 doesn't point any more to the GRIB message it was supposed to.
Please revise your code in view of this example or link to the new version and add a call
to the GRIB_RELEASE function to release the memory when the which is not any more released in the new version.
* Second order packing algorithm rewritten and working in GRIB edition 1 and 2 with
packingType=grid_second_order.
* This version is ready to be used with the new GRIB1 and GRIB2 interpolation function (INTF2)
available in EMOSLIB.
* A significant change in the fortran interface has been done for which some software using the
previous versions could exhibits memory allocation problems due to unreleased memory associated
with the GRIB messages. In the previous versions (starting from version 1.6.1) the argument IGRIB
of the function GRIB_NEW_FROM_FILE(IFILE,IGRIB) was an input and output argument.
In the new version is output only and the grib message associated with IGRIB will never be released
by the GRIB_NEW_FROM_FILE function. This means that to release the memory associated to the message
a call to GRIB_RELEASE(IGRIB) has to be introduced when the message is not needed any more. In the
previous version the memory was released when a value of IGRIB associated with a valid message was
passed to the function. It has been realised that this behaviour was source of unexpected results
when the IGRIB variables used in the user code were not initialised to a non valid IGRIB value
(zero or negative integer).
* Review and update of the code tables.
* Added new parameters. Full list of parameters accessible from
http://www.ecmwf.int/publications/manuals/d/gribapi/param/
* Bug fixes.
14-10-2010 Enrico Fucile
-- released version 1.9.5
* Python interface available with the configure option --enable-python.
See http://www.ecmwf.int/publications/manuals/grib_api/namespacegribapi.html
and http://www.ecmwf.int/publications/manuals/grib_api/grib_examples.html
* Available new environment variable GRIB_API_IO_BUFFER_SIZE defining the size in bytes of the buffer
used in the IO calls from Fortran and in the tools.
* Performance improvements and partial vectorisation for the NEC SX platform.
* Review and update of the code tables.
* Update of paramId, shortName, name key values and release of a new parameter database web interface.
* Added a new key called 'global' for regular_gg and reduced_gg which can take the values 1 ( =True) or 0 (=False).
This is used to specify whether the grid is global or not. Setting global=1 will set the grid parameters computed
only from N and Ni. It is used in the conversion from one edition to the other.
* Available new environment variable GRIB_API_NO_ABORT. Default value is 0 and when it is set to 1 the library will not
abort on failing asserts.
* Bug fixes.
07-05-2010 Enrico Fucile
-- released version 1.9.0
* Setting paramId or converting from one edition to the other when a parameter is not
defined in the paramId.def file will result in a failure.
* Improved performance in reading/cloning a message.
* Changed the indexing system and provided new functions to write the index on a file
grib_index_write and read it grib_index_read
* Provided a new tool to build an index from a set of files: grib_index_build.
* The tool grib_compare is able to compare two index files, going through the grib messages
indexed in the first and searching for them in the second.
* definition files and library have been updated to be used in operational status in the
IFS cycle 36R2
* grib_ls has slightly changed format.
* A -j option is provided in grib_ls (-l only) to have the output in json format.
* BUFR messages are read, but only very basic decoding is provided.
* Parameters have been updated.
* Native type of paramId has been changed from string to integer.
* Several small bug fixes.
23-07-2009 Enrico Fucile
-- released version 1.8.0
* Improved conversion from edition 1 to 2. Set edition=2 will do the conversion
from edition 1 to 2 also of the parameter information.
* shortName, paramId and units are defined for all the official WMO parameters
and for local parameters used at ECMWF and NCEP
* grib_compare has been modified to compare grib 1 and 2 (option -e).
Other important changes have been done on grib_compare to have a more powerful tool to compare headers and
values. New options are provided to choose the tolerance and new namespaces are available to
compare only the relevant part of information. The new namespaces are.
parameter,geography,vertical,time,local,data.
* For some ECMWF local definitions the conversion rules have been included in grib_api.
* A new key cfName is provided for the netcdf CF name. It is at the moment defined
only for some parameters.
* The parameter information is generated from a parameter database maintained at ECMWF
and accessible from a new section in the grib_api web pages.
* A new key typeOfLevel valid for grib edition 1 and 2, with the following values has been introduced
'surface', 'cloudBase', 'cloudTop', 'isothermZero', 'adiabaticCondensation', 'maxWind',
'tropopause', 'nominalTop', 'seaBottom', 'isobaricInhPa', 'isobaricInPa', 'isobaricLayer',
'meanSea', 'isobaricLayerHighPrecision', 'isobaricLayerMixedPrecision', 'heightAboveSea',
'heightAboveSeaLayer', 'heightAboveGroundHighPrecision', 'heightAboveGround', 'heightAboveGroundLayer'
'sigma', 'sigmaLayer', 'sigmaLayerHighPrecision', 'hybrid', 'hybridLayer', 'depthBelowLand',
'depthBelowLandLayer', 'theta', 'thetaLayer', 'pressureFromGround', 'pressureFromGroundLayer',
'potentialVorticity', 'depthBelowSea', 'entireAtmosphere', 'entireOcean', 'oceanWave', 'oceanMixedLayer'
* The open source library OpenJpeg can be used in place of Jasper for the jpeg 2000 coding.
--with-openjpeg options in the configure.
* decimalPrecision can be used also with grid_jpeg packing.
* parameterName,parameterUnits are provided from the code tables.
* packingError is also available for spherical harmonics packing. It provieds the error due to simple packing.
* option -l added in grib_get
* option -F added in grib_get and grib_ls to control the format of floating point keys.
* grib_ls output has been changed due to the introduction of the new typeOfLevel key
* Subroutine grib_is_missing added in the Fortran interface to check if a header value is missing.
* Changed step behaviour. If the step is "instant" and the value doesn't fit into one byte (in grib 1)
* an appropriate value of timeRangeIndicator is found to code the value in two bytes instead of 1.
* In grib_filter the function "append" has been added to append a message to a file as opposed to "write" which is
writing from the beginning of the file.
* A new grib_count function is available to count (very fate) the number of messages in a list of files.
* The value of the key level is zero in both edition when it is not relevant for the message and it is
missing in grib edition 2.
* The indexing routines have been improved to process string keys.
06-03-2009 Enrico Fucile
-- released version 1.7.0
* New indexing routines.
Example: index.f90
(http://www.ecmwf.int/publications/manuals/grib_api/grib_examples.html)
* New keys stepType,stepUnits,stepRange,startStep,endStep.
* New keys swapScanningX, swapScanningY, swapScanningLat, swapScanningLon. To
swap the field respect one axis. They swap data and fix the header accordingly.
* "Improved" error messages.
* write multi fields. Previous versions of grib_api were able only to read multi
fields, now you can also write multi fields. We don't suggest to you this
feature, it is included only to be compliant with FM-92 spec.
Examples: multi.f90, multi_write.f90
(http://www.ecmwf.int/publications/manuals/grib_api/grib_examples.html)
* function grib_new_from_template is deprecated and it will be replaced with
grib_new_from_samples. Also the environment variable GRIB_TEMPLATES_PATH is
replaced by GRIB_SAMPLES_PATH. Please replace "template" with "samples" in all
the C functions also. The behaviour is the same. I made a new example showing the
best practise to use samples. The technique of cloning is strongly recommended.
Example: samples.f90
(http://www.ecmwf.int/publications/manuals/grib_api/grib_examples.html)
* New key decimalPrecision provided. Used in the previous example. Regarding
decimalPrecision and bitsPerValue the following problem was found in the
previous version. If after loading from file or from sample (template) or
cloning a constant field you set some non constant values the library doesn't
know the bitsPerValue or the decimalPrecision you want to code (those are zero
in the constant field). The result could have been a field coded with a poor
precision. Starting with the new 1.7.0 version in these ambiguous cases grib_api
is using 24 bits to pack the field (safe value). We always suggest to set
explicitly the bitsPerValue or decimalPrecision or to clone non constant fields
with those values properly set.
* grib_copy, grib_set. A key name in square brackets can be used in the output
file name. The key will be replaced with its value in the processed grib
message. This will provide an easy way for splitting fields as the file name
will be dependent on some key values. Example:
grib_copy input.grib "output_[shortName]_[date]_[step].grib"
Please remember to quote the file name otherwise you can get strange behaviours
due to the shell.
* New key packingError. The error introduced by the packing algorithm is
returned with this key. It is available only for some packing algorithms. All
the packed values are affected by this error in the sense that
original value before packing = unpacked value ± packingError
packingError is different for different fields (also for different steps of the
same parameter/level/run).
* all the nearest functions are returning the distance from the grid point in
Km, using the radius coded in the message.
* key bitmapPresent now available also in grib 2. Example: bitmap.f90
(http://www.ecmwf.int/publications/manuals/grib_api/grib_examples.html)
30-09-2008 Enrico Fucile
-- released version 1.6.4
* Fields containing a bitmap with all missing values can be coded in grib1 and grib2
* Multifields grib2 messages can be loaded from memory.
* local definition 31 added
* Better use of memory
* minor bug fixes
16-07-2008 Enrico Fucile
-- released version 1.6.1
* More efficient use of memory and improvement of performance.
* Thread safe version available through the --enable-pthread configure option.
* New computation of kindOfInt, kindOfLong, kindOfDouble, kindOfFloat, kindOfSize
to allow safe arguments handing between Fortran 90 and C.
* A new key bitmap is provided to get the full bitmap as a string with the grib_filter
instruction: print "[bitmap]";
* all the tools are failing on error and can be forced to proceed execution with
the option -f . The only exception is grib_ls because we already have an identical
tool failing (grib_get).
* new error messages from the Fortran 90 interface will give you some more chance to debug
your code.
* New function grib_count_in_file available from Fortran and C to count the messages in a file.
See example: examples/F90/count_messages.f90
* In the Fortran 90 interface any valid grib id passed to a grib_new_from_file or grib_new_from_message
will be released before being assigned to a new grib message.
Example
call grib_new_from_file(ifile,igrib, iret)
do while (iret /= GRIB_END_OF_FILE)
... do something with the message pointed by igrib
... igrib will have the same value, but
... it will point to a different message each cycle
... the message from the previous cycle will be released
... no need for grib_release(igrib)
call grib_new_from_file(ifile,igrib, iret)
enddo
If you want to keep the grib message in memory remember to pass a negative grib id.
Example
call grib_count_in_file(ifile,n)
allocate (igrib(n))
do i=1,n
call grib_new_from_file(ifile,igrib(i), iret)
...
enddo
... remember to release the grib ids to free memory
do i=1,n
call grib_release(igrib(i))
enddo
* grib_filter is now able to print with a format, number of colums and a separator for array keys.
The format is expressed with a % followed by a C like (printf) format declarator.
The number of columns is started with !
The separator string is delimited by ' '
Examples
print "[values!6%.5f',']" values with 5 digits precision in 6 columns separated by ,
print "[latitudes%.3f!1]" latitudes with 3 digits precision in a column
print "pl={[pl!7', ']}" pl (number of point for each latitude) in 7 columns separated by ', '
* New keys:
julianDay
latLonValues array containing latitude(1),longitude(1),value(1), latitude(2),longitude(2),value(2),...,
latitude(N),longitude(N),value(N) (N=number of points in the grid)
latitudes latitude(1),latitude(2),...,latitude(N) (N=number of points in the grid)
longitudes longitude(1),longitude(2),...,longitude(N) (N=number of points in the grid)
distinctLatitudes list of distinct latitudes
distinctLongitudes list of distinct longitudes
* minor bug fixes
03-06-2008 Enrico Fucile
-- released version 1.5.0
* Review of packing algorithms. Improvement of performance mainly on big endian
platforms.
* Modified the internal retrieval of keys, improving performance in getting and setting
keys.
* Implemented memory management. An efficient tecnique to reuse memory is implemented.
--disable-memory-management configure option available to disable the memory management
and use system malloc/free.
* Some experimental multithreaded packing is available for big endian machines.
A configure option --enable-omp-packing is provided to enable the openMP instructions
contained in the new packing. The environment variable OMP_NUM_THREADS sets the number
of threads used in the packing algorithm.
* grib_filter improvement:
* it is now possible to set array keys like values or pl as follows:
values = { 0.1, 3.2, 7.5 };
* it is now possible to print array keys like values or pl as follows:
print "values=[values]";
it is also possible to print on a file ("test.out") doing:
print("test.out") "values=[values]";
* better error messages are printed
* it does not fail if an undefined key is used in a write or print statement
* New keys or changed keys:
* a scale factor and an offset can be applied to the data values through the keys
"scaleValuesBy" and "offsetValuesBy".
* setBitsPerValue is the new key to be used to change the number of bits per value.
It also executes the repacking.
* The keys periodOfTime is now working also when indicatorOfUnitOfTimeRange=0 (time units is
minutes)
* Concepts can be defined by the user. Concepts are a way to link the value of a key to
a combination of values of other keys. They are used to define the short_name and the
name.
* minor bug fixes
07-02-2008 Enrico Fucile
-- released version 1.4.0
* New Fortan 90 interface. Main featues:
* unique grib_get/grib_set subroutines to get/set any variable type.
* status code is an optional argument in all the subroutines (when omitted
the program will exit in case of error)
* new subroutine grib_get_data to get the latitude/longitude/values arrays.
* new subroutine grib_get_element to get one or several elements of an array key
through their index.
* new subroutine grib_nearest function to get the nearest grid point of a given lat/lon.
It accepts in input also arrays of latitude/longitudes to find the nearest points in one
single call. A land sea mask mode is also available to get the nearest land point.
* New Fortran 90 examples and documentation.
* New grib_keys tool to obtain the list of keys available for a specified type of grib.
* Modified the -l option in grib_ls (see documentation or type grib_ls without arguments)
* fixed a bug in the grib_filter and added the -o option for the output file.
* minor bug fixes.
12-11-2007 Enrico Fucile
-- released version 1.3.0
* grib_ls -l lat/lon/mode to retrieve values in the nearest grid points of a given
latitude longitude point
* grib_ls -B"order by" to order the output using some keys
* grib_ls -i index. To extract the value in the index grid point.
* new grib_dump exposing a minimum set of keys which are the ones suggested for the
standard use. Some options added to the grib_dump. The output of this new
grib_dump is ready to be coupled to the new grib_gen which will be released in the
next version.
* grib_copy -r to repack data fixing problems in the original file
* improved keys search for faster tools
* grib_get_data and grib_iterator working with all scanning mode
* new experimental functions introduced, not available in the fortran
interface: grib_fieldset_new,grib_fieldset_apply_order_by,grib_fieldset_next_handle
grib_nearest_new,grib_nearest_find,grib_nearest_delete
* some examples on the grib_fieldset and grib_nearest are added.
* The management of the missing data in grib_get_data is changed. Please refer to the help
obtained executing grib_get_data without arguments.
* Added the key stepRange which is native type string. It can represent start and end of a
cumulation or a max of min interval. Example stepRange=24-36 for a precipitation means that it
is cumulated between 24 and 36. It also addresses the problems with the big steps which don't
fit in the grib1 P1 and P2.
* Added key numberOfCodedValues. It is different from numberOfValues only if a bitmap is present and
the difference between the two should be numberOfMissingValues.
* Added the option -g to grib_copy, grib_set, grib_convert, grib_filter to preserve GTS headers
wrapping the grib message
* fixed some bugs in the fortran interface
* fixed a bug in reading the signed floats with the grib_convert.
01-09-2007 Enrico Fucile
-- released version 1.2.0
* changed license. Now grib_api is distributed under LGPL see LICENSE file in distribution.
* changed io functions and configure.ac to support files larger than 2GB
* as a consequence to the large file support the key offset has native type double
* added support for predefined grids
* modified grib_handle_new_from_file to read messages with wrong message length.
In this case an error will be raised.
* A new -M option is provided for all the tools to turn off the multi-field grib
support.
* Changed behaviour of grib_copy. Default is quiet and option -v (verbose) is available.
09-07-2007 Enrico Fucile
-- released version 1.1.0
* fixed a bug affecting constant fields.
* rewritten the ibm floating point encoding/decoding routines gaining a factor of 7 in performance
* fixed a problem in the configure affecting compilation of fortran programs on AIX platforms
* Added some new keys:
* max,min,average (maximum, minimum and average respectively of the data )
* numberOfDataPoints(alias numberOfPoints) computed from the grid description
* numberOfValues is the same as the numberOfPoints if a bitmap is not present
otherwise it takes into account the bitmap
* numberOfMissing (alias numberOfMissingValues) number of missing values in the field
It's easy to check if a field is constant by simply comparing max,min,average
and some information about bitmap and missing values can be found comparing numberOfValues
numberOfPoints, numberOfMissing.
* added some new tests
* added a new example on how to print all the data when the iterator is not available.
* grib_tools are now failing with an error when the input file does not contain any
valid grib message
* added management of reduced gaussian non global fields.
17-04-2007 Enrico Fucile
* grib_api version 1.0 released