Detailed description of changes included within each new release.

Version 11.7.1 - August 26, 2022

More extensions were added to support the query interface for C++ IODA converters. [Issue #225]
Several internal routines were modified to allow the encoding and decoding of values larger than 32 bits. [Issue #195]

Version 11.7.0 - May 19, 2022

Extensions were added to support a new query interface for C++ IODA converters. [Issue #198]
Updates and bug fixes were made to several utility programs within the library package. [Issue #141]
A fix was made to prevent a line truncation error which occurs with certain compilers when the value of the MASTER_TABLE_DIR macro exceeds a certain length. [Issue #182]

Version 11.6.0 - November 10, 2021

All of the library builds now use dynamic allocation, so the "_DA" suffix has now been correspondingly removed from the names of the library builds. [Issue #77]
The default BUFR master table version number for output BUFR messages was changed from 29 to 36 within subroutines cmsgini(), dxmini() and msgini(). [Issue #142]
The maximum length of a [path/]filename that can be passed into function cobfl() was increased to 200 characters. [Issue #174]
Subroutine ufbpos() was modified to allow it to work for all types of BUFR messages, including compressed messages. [Issue #170]
The library was modified to read up to 24 characters from the units field of BUFR master Table B files. Furthermore, in accordance with updated guidance from WMO, all fields within all BUFR master table files are now restricted to the CCITTIA5 character set. [Issue #140]
Subroutine jstchr and function rjust have been removed from the library, and their usage has been replaced with the Fortran intrinsic functions adjustl and adjustr. [Issue #165]
Several prototypes were added to the bufrlib.h header file to enable further interoperability with C application programs. In addition, several existing prototypes were modified to use size_t when passing string lengths between Fortran and C. [Issue #164]

Version 11.5.0 - April 26, 2021

Subroutine ufbseq() was modified for cases where the number of available levels exceeds the amount of user-provided array space when reading from an input file. Previously, the software would abort in such cases without returning any data whatsoever, but with the change it will now print a diagnostic and return with the number of levels equal to the amount of array space that was provided. [Issue #52]
Subroutine ufdump() was modified to improve the logic for sequence tracking, and to increase the format width for integer code figure and bit number values when printing associated meaning strings. [Issue #55]
Subroutine ufbdmp() was modified to fix a bug when checking for the "missing" value in long character strings (i.e. longer than 8 bytes) and to enable printing of up to 120 characters for such strings. [Issue #55]
Subroutine readlc() was modified to return a "missing" character string if the requested mnemonic isn't found in the subset. Previously, the subroutine would return a string of all blank characters in such cases. [Issue #53]
Subroutine writlc() was modified to print a warning message if the requested mnemonic isn't found in the subset definition. The library would previously abort in such cases. [Issue #53]
The default directory location of the BUFR master tables in subroutine bfrini() was changed from a fixed WCOSS-specific path. It is now set to the defined value of the MASTER_TABLE_DIR macro when the library was built. [Issue #123]
Subroutine writsa() was modified to allow the return of up to two BUFR messages during the same call to the subroutine, in the rare instances where more than one BUFR message could become available during such a call. [Issue #54]
Several internal Fortran variable declarations were modified for compatibility with GNU v10+ compilers. [Issue #81]
Subroutines pkx and chrtrn are no longer used within the library and were never intended to be called from application codes, so they have now been removed from the library. [Issue #107]

Version 11.4.0 - November 20, 2020

A Python API was added to the library, for use with Python applications. [Issue #61]

Version 11.3.2 - July 16, 2020

A user-friendly sanity check was added to subroutine closbf(), in case an application program is using a dynamic-allocation build of the library and calls this subroutine prior to calling subroutine openbf(). The library would previously abort in such cases, but it will now just print a warning message. [Issue #9]

Version 11.3.1 - March 3, 2020

Subroutine stndrd() was patched to fix an internal calculation for messages containing only one subset. [Issue #51]

Version 11.3.0 - May 21, 2019

A bug was fixed in the bitmap processing, so that each Table A mnemonic in the jump/link table now tracks its own individual set of underlying Table C operators. This involved changes to function igetrfel() and subroutines tabsub() and makestab().
The build script makebufrlib.sh was modified to remove calls to cpp (the C preprocessor) for all *.F Fortran files. Preprocessing for these files is now done by fpp (the Fortran preprocessor), which is automatically called by the Fortran compiler for all *.F files.
Global variable MXRST (the maximum number of long character strings that can be read from a subset of a compressed BUFR message) has now been added to the list of array size limits that can be configured using dynamic allocation.
The build script makebufrlib.sh was modified to stop generating the s_64 ("supersize") build. Such builds are now obsolete, since users can use the dynamic allocation builds to flexibly define whatever size limits they need for large BUFR datasets. If users need to recreate the exact specifications that were present in the previous s_64 build, they can do so using a dynamic allocation build and separate calls to subroutine isetprm() to set MXLCC=12, MAXMEM=75000000, MXCDV=50000, MAXJL=128000, MAXSS=150000, MXMSGL=2500000 and MXRST=500 within their application program.
A new capability was added to read and process master code and flag tables. This feature is activated via an initial call to new subroutine codflg() at any time after the first call to subroutine openbf(). Once that is done, then for any BUFR message read into the library via subroutine readmg(), readerme() or equivalent, and for any mnemonic contained within that message which is defined as a code or flag table, the user may call new subroutine getcfmng() with that mnemonic and an associated value (code figure or bit number) as input, and the subroutine will return the meaning (as a character string) corresponding to that mnemonic and associated value.
The default BUFR master table version number was changed from "13" to "29" within subroutines cmsgini(), dxmini() and msgini().
The maximum number of delayed replication factors that can be passed in as input to subroutine drfini() was increased from 200 to 2000.
Subroutines msgupd() and cpyupd() were modified so that the IPRT verbosity must be greater than or equal to 1 in order for diagnostic alerts to be printed whenever BUFR subsets greater than 65535 bytes are written to their own output messages.
Function i4dy() was modified to change the window for converting 2-digit years to 4-digit years. The old window was 1921-2020, i.e. add 1900 to any 2-digit year greater than 20; otherwise add 2000. The new window will be 1941-2040, i.e. add 1900 to any 2-digit year greater than 40; otherwise add 2000.

Version 11.2.0 - April 11, 2017

Support was added for the processing of 2-2X-255, 2-3X-255 and 2-4X-255 marker operators when reading BUFR messages, including when such operators are included in sequences within a BUFR DX table.
A new subroutine gettagre() was added which, given an element within a subset open for reading, determines if the element references another element within the same subset via an internal bitmap, and if so returns the referenced element along with its location in the subset.
A new function igetmxby() was added which returns the maximum size of a BUFR message that can be written to any output stream by the BUFRLIB software. This value can then be modified via a subsequent call to subroutine maxout().
Subroutine nemtbd() was modified to allow the last mnemonic in a sequence to be a "following value" mnemonic, for cases such as when it is being used as a coordinate descriptor and set to a value of "missing" to cancel a previous instance of the same descriptor.
A new subroutine setvalnb() was added which searches for a specified mnemonic in a subset definition, then searches forward or backward from that point for a different mnemonic and sets the associated value as specified by the user. It can be useful in certain application codes which write BUFR output.
Subroutines rdtree() and rcstpl() were modified to automatically identify certain cases where a decoded subset is corrupt and return this information to the application program via readsb() or ireadsb(), rather than continuing to try to decode the message which could in turn lead to a segmentation fault.
The build script makebufrlib.sh was updated to add a missing compiler option needed for compatibility with the Cray programming environment on the s_64 ("supersize") build.
A bug was fixed in subroutine ufbtab() which could occasionally lead to array maximums being exceeded and corrupted output when reading BUFR files.

Version 11.1.0 - April 27, 2016

Subroutine wrcmps() was modified to fix a bug involving the encoding of long character strings (via subroutine writlc()) into compressed messages which also contain delayed replication sequences.
Subroutine msgupd() was modified to always call subroutine usrtpl(), even for overlarge subsets which don''t get written to the output stream. This ensures that such subsets are properly flushed from internal arrays.

Version 11.0.2

The build script makebufrlib.sh was updated for compatibility with the Cray programming environment, along with module MODA_MSTABS and functions icbfms() and isize(). All changes remain compatible with other supported environments.

Version 11.0.1

Subroutines cpyupd(), ufbmem(), ufbmex() and ufbovr() were patched to include the proper declaration for IPRT, which is a global variable controlling the verbosity of diagnostic output.
Subroutine ufdump() was modified to include level identifiers for event stacks in the print output.
Subroutine wrcmps() was modified to fix a bug involving a variable that was not being saved between successive calls to this subroutine and which in rare cases could result in the loss of output subsets.

Version 11.0.0 - April 27, 2015

A new subroutine rtrcptb() was added which works just like subroutine rtrcpt(), except that it operates on a BUFR message passed directly to it by a call argument, rather than on the last BUFR message that was indirectly read during the previous call to subroutine readmg(), readmm(), readerme() or equivalent. Subroutine rtrcpt() was correspondingly modified to now directly call rtrcptb(), to avoid maintaining the same program logic within two different subroutines.
The maximum number of delayed replication factors that can be passed in as input to subroutine drfini() was increased from 100 to 200.
Subroutine ufbseq() was modified to fix a bug involving nesting of delayed replication sequences. The subroutine would fail to store data properly in cases where the inner-nested sequence was not present (i.e. zero replications) within the first replication of the outer sequence.
The makebufrlib.sh build script was modified to automatically extract the version number for the current build from the source of subroutine bvers(). Previously this information had been hardcoded in multiple lines of the script.
The default directory location of the BUFR master tables in subroutine bfrini() was changed from "/nwprod/fix" to "/nwprod/decoders/decod_shared/fix".
A new subroutine nemspecs() was added which returns the scale factor, reference value and bit width corresponding to a specified occurrence of a given mnemonic within a subset definition, including accounting for any Table C operators such as 2-01-YYY, 2-02-YYY, 2-03-YYY, 2-07-YYY, 2-08-YYY, etc. which may be in effect for that particular occurrence of the mnemonic.
A new subroutine nemdefs() was added which returns the element definition and units associated with a given mnemonic.
Global parameter MXIMB was removed from the BUFRLIB, and the corresponding logic within subroutine mvb() was simplified to mitigate the need for it.
Subroutines msgupd(), cpyupd() and copysb() were modified to ensure that any subset larger than 65530 bytes is written to its own message in the output stream. This ensures that the subset byte count indicator, which is an NCEP local descriptor packed into 16 bits prior to each subset in a message, will not need to be relied upon in order to locate any subsequent subsets within the same message.
Global parameter MAXSS (the maximum number of data values that can be read from or written into a subset by the BUFRLIB software) was increased from 80000 to 120000.
Subroutine RDMSGB, which was rendered obsolete with the addition of the embedded C-language I/O upgrade in version 10.2.0, has been removed.
A logical call argument was added to subroutine upc() to let the calling subprogram specify whether null characters should be converted to blanks. Previously, all null characters were converted to blanks by default.
New versions of all of the normal builds of BUFRLIB have been added which allow certain array sizes to be specified by the user at run time, with the corresponding arrays dynamically allocated at run time rather than statically allocated at compile time. The new builds contain the suffix "_DA" and can be used by application programs which need flexibility in defining size limits for certain outlier BUFR datasets. The size limits that can be modified are defined within new subroutine isetprm(), which must itself be called prior to the first call to subroutine openbf() for each new size limit that is to be modified from its default value. The corresponding arrays are then dynamically allocated during the subsequent first call to openbf(). As part of this enhancement, numerous subprograms within BUFRLIB have been rewritten to share memory using FORTRAN modules rather than common blocks, and new conditional compilation flags have been incorporated into these subprograms as well as to the makebufrlib.sh build script. This allows the same source code and build script to be used for both dynamic allocation and static allocation builds. Note that application programs which don't have a need to redefine any default array size limits may continue to use the existing static allocation builds for maximum runtime efficiency.
A new subroutine pkx() was added which works just like subroutine pkb(), except that it properly handles cases where the input value NBITS is greater than the number of bits in a machine word.
Support was added for the processing of 2-2X, 2-3X and 2-4X non-marker operators when reading or writing BUFR messages, including when these operators are included in sequences within a BUFR DX table.
Subroutine stseq() was modified to fix a bug involving the application of associated fields to Table D sequence descriptors.
Function icbfms() was modified to improve the logic for identifying "missing" strings encoded as REAL*8 10E10 values prior to version 10.2.0 of the library.

Version 10.2.5

Subroutine mesgbf() was modified to ensure that the input BUFR file is always closed before exiting the subroutine.
Function cobfl() was modified to allow up to 500 characters in the path of the filename being opened.
A declaration typo was fixed in subroutine blocks().
Global parameter MAXNC (the maximum number of FXY descriptors that can be written into Section 3 of a BUFR message) was increased from 300 to 600.

Version 10.2.4

Configuration files bufrlib.PRM and makebufrlib.sh were updated to generate a 4_32 build (4-byte REAL, 4-byte INT, 32-bit compilation) on the IBM CCS for version 10.2.3 of the BUFRLIB.

Version 10.2.3 - June 21, 2013

Subroutine rdusdx() was modified to prevent a segfault when trying to read DX dictionary information from an empty file.

Version 10.2.2

Subroutine openbf() was modified to fix a bug which caused a segfault in certain cases when appending to a BUFR file using the embedded C-language I/O.
Subroutines readlc() and writlc() were modified to allow the input mnemonic string to be up to 14 characters when it contains a '#' condition code.

Version 10.2.1

A bug was fixed in the embedded C-language I/O to account for the difference in index numbering between Fortran and C arrays.

Version 10.2.0 - October 19, 2012

The makebufrlib.sh script was modified to streamline the endianness check and make it more portable.
Subroutine wrtree() was modified to ensure that "missing" character strings are properly encoded with all bits set to 1.
A new function icbfms() was added which tests whether decoded character strings are "missing" by checking if all of the equivalent bits are set to 1. This was done because, on certain platforms, the BUFRLIB REAL*8 "missing" value BMISS is not always equivalent to all bits set to 1 when viewed as a character string, and thus the existing BUFRLIB function ibfms() did not always work properly in such cases. However, users can continue to use the existing IBFMS function in application programs, because the new icbfms() function has now been incorporated internally within the logic of many BUFRLIB subroutines, in addition to also being available for direct calling by application programs.
Subroutines readmg() and readerme() were modified to prevent the BUFRLIB from internally adjusting to DX (dictionary) table messages when Section 3 decoding is being used. Otherwise, contention can occur between the table information in the DX messages and the table information specified within the Section 3 descriptors. From now on, whenever Section 3 decoding is used (as specified by setting IO="SEC3" when opening a file via openbf()), the BUFRLIB will now treat any DX (dictionary) table message the same as any other message and decode the actual data (i.e. table) values according to Section 3.
Subroutine openbf() was modified to allow a new option for input call argument IO. If this argument is set to 'INUL', then the BUFRLIB will behave the same as when IO='IN', except that it will never try to actually read anything from the file attached to input call argument LUNIT. This can be useful for some special cases, such as when the user plans to pass input messages to the BUFRLIB using subsequent calls to subroutine readerme().
A new subroutine gettagpr() was added which returns the mnemonic corresponding to a parent sequence in a subset definition, given the mnemonic corresponding to a child descriptor within that sequence. This can be useful in certain application codes, especially when Section 3 decoding is being used.
A new function getvalnb() was added which searches for a specified mnemonic in a subset definition, then searches forward or backward from that point for a different mnemonic and returns the associated value. This can be useful in certain application codes, especially when Section 3 decoding is being used.
Functionality was added to improve the portability of reading and writing BUFR messages across different platforms. All calls to existing FORTRAN subroutines which read or write BUFR messages from disk (e.g. readmg(), ufbmem(), ufbtab(), writsb(), wrcmps(), copymg(), etc.) now use embedded C-language I/O to perform these tasks. Among other things, this means that any BUFR file can now be read regardless of whether it has been pre-blocked with FORTRAN control words using the cwordsh utility. For writing BUFR files, a new subroutine setblock() was added which allows users to specify whether output BUFR messages are to be unblocked (which is the new default), big-endian blocked, or little-endian blocked.
A new subroutine setbmiss() was added which allows users to specify a custom "missing" value for writing to and reading from BUFR files, rather than using the BUFRLIB default "missing" value of 10E10. A corresponding function getbmiss() was also added which returns the current "missing" value in use.

Version 10.1.0 - June 11, 2012

Subroutine ufdump() was modified to fix a bug when checking for the "missing" value in long character strings (i.e. longer than 8 bytes).
A new subroutine ufbmex() was added for use with certain application programs. ufbmex() functions similarly to ufbmem(), but has an additional return argument containing an array of message types corresponding to the array of messages that were read into internal memory.
Subroutines ADDATE, IUPBS1, IUPVS1, LJUST, LSTRPC, LSTRPS, SUBUPD, POSAPN and PARSEQ, which had been marked as obsolete within a previous version of BUFRLIB, have now been deleted.
Several global parameters were increased in "bufrlib.PRM". Specifically, MAXTBA (the maximum number of Table A entries for a BUFR file) was increased from 120 to 150, and MXDXTS (the maximum number of dictionary tables that can be stored for use with BUFR messages in internal memory) was increased from 10 to 200.
Subroutine conwin() was modified to fix a bug and remove an obsolete call argument.
Subroutine wrcmps() was modified to fix a bug involving embedded tables within a file of compressed BUFR messages.
Documentation was improved in many subroutines throughout the library.
Support has been added for the 2-03-YYY "change reference values" operator.
Subroutine usrtpl() was modified to fix a bug that was incorrectly using parameter MAXJL instead of parameter MAXSS when checking for overflow of an internal template array.
Subroutine wrdxtb() was modified to prevent it from trying to store more than 255 Table A, Table B or Table D descriptors in a single DX dictionary message. The maximum value was set to 255 since regular 8-bit delayed replication is used to store descriptor information in these messages.
Subroutine tabsub() was modified to correctly generate the jump/link table for subsets where a Table C operator immediately follows a Table D sequence.

Version 10.0.1

Subroutine rewnbf() was modified to fix a bug which skipped the first data message after a file rewind.

Version 10.0.0 - August 12, 2010

Subroutines PKVS1, OVRBS1, NMBYT, READIBM, IREADIBM, READFT, IREADFT and MOVA2I, which had been marked as obsolete within a previous version of BUFRLIB, have now been deleted.
A new global parameter MAXSS was defined for use as the maximum number of data values that can be read from or written into a single data subset by the BUFRLIB software. Previously, the separate global parameter MAXJL was used to define this limit. MAXJL will now be used solely to define the maximum number of internal jump/link table entries.
The size of a string declaration was increased within subroutine rdusdx().
Subroutine readlc() was modified to enable the extraction of "long" (i.e. greater than 8 bytes) character strings from compressed messages. In addition, it is now possible to access all occurrences of such a string from within a given data subset, via the use of the new mnemonic condition character '#'. Previously, readlc() could only ever access the first occurrence of any "long" character string from within a data subset.
Subroutine writlc() was modified to allow the writing of "long" (i.e. greater than 8 bytes) character strings within compressed messages. In addition, it is now possible to write all occurrences of such a string into a given data subset, via the use of the new mnemonic condition character '#'. Previously, writlc() could only ever locate and write the first occurrence of any "long" character string within a data subset.
Subroutine ufdump() was modified to label each output level for sequences where the replication count is greater than 1. In addition, it will now output all occurrences of "long" (i.e. greater than 8 bytes) character strings from within a given data subset.
Subroutine rdcmps() was modified to fix a bug which could cause the overflow of internal arrays when working with long character strings (i.e. longer than 8 bytes).
Subroutine nvnwin() was modified to fix a bug which could cause the overflow of an internal array during initialization on certain operating systems.
A new subroutine bvers() was added as a resource for managing and reporting library version numbers.
The fuzziness threshold in function ibfms() was increased for improved accuracy when testing for the BUFRLIB "missing" value.
A new subroutine iupbs3() was added which unpacks specified values from Section 3, including subset counts and compression indicators. The same logic had been repeated within numerous existing subroutines throughout BUFRLIB and has now been consolidated into this single subroutine that can itself be called from wherever it is needed.
Subroutines readerme(), rdmsgw() and RDMSGB were modified to prevent the overflow of an internal array for extremely large BUFR messages.
Subroutine upds3() was modified to pass in a new input argument containing the dimensioned size of the output array, in order to prevent the subroutine from possibly overflowing the array.
Subroutine writsa() was modified to pass in a new input argument containing the dimensioned size of the output array, in order to prevent the subroutine from possibly overflowing the array.
A new capability was added to BUFRLIB to enable the decoding of a BUFR message according to its data description section (Section 3). This is activated by setting IO="SEC3" when opening the file via subroutine openbf(). Master tables containing all possible BUFR descriptors are also required, and these may be specified via a call to new subroutine mtinfo() or by using default values specified within subroutine bfrini(). If the default values are used, then FORTRAN logical unit numbers 98 and 99 will be allocated by the BUFRLIB for opening and reading the master tables. This new capability allows BUFR messages to be decoded without pre-defined DX dictionary files.
Subroutine readmm() was re-written to directly call subroutine rdmemm() instead of duplicating all of the code logic in rdmemm().
Subroutine upb() was re-written to directly call subroutine upbb() instead of duplicating all of the code logic in upbb().
Subroutine POSAPN has been marked as obsolete (for future removal from BUFRLIB). The same functionality can now be obtained via the use of subroutine posapx().
Subroutine wrcmps() was modified to fix a bug involving the writing of compressed subsets which contain delayed replication. In certain situations, the values of two internal variables were not being properly saved between successive calls to the subroutine.
Changes were made so that the BUFRLIB will automatically read and adjust to any DX table (dictionary) messages internal to a file. Previously, the software would only ever process such messages at the beginning of a file, so that all subsequent data messages in that file were required to conform to these initial dictionary messages, and any subsequent dictionary messages in the file were simply ignored. Now, any subsequent dictionary messages will cause the BUFRLIB to adjust its internal processing tables and treat all subsequent data messages as conforming to these new dictionary messages, up through the end of the file or until yet another set of dictionary messages is encountered. These changes affect all BUFRLIB subroutines which read BUFR messages from a file, including readmg(), ireadmg(), readmm(), ireadmm(), rdmemm(), readns() and ireadns().
Subroutine ADDATE has been marked as obsolete (for future removal from BUFRLIB) since it is no longer called by any BUFRLIB routines. The same functionality can now be obtained via the use of subroutine W3MOVDAT in the NCEP W3 library.
Subroutine SUBUPD has been marked as obsolete (for future removal from BUFRLIB) since it is no longer called by any BUFRLIB routines and is almost an exact replica of subroutine msgupd(). The same functionality can now be obtained via the use of subroutine msgupd().
A new logical function msgfull() was added which determines whether there is enough room to store the current data subset within the current BUFR message for output. The same logic had been repeated within numerous existing subroutines throughout BUFRLIB and has now been consolidated into this single subroutine that can itself be called from wherever it is needed.
A new capability was added to BUFRLIB to allow it to append a tank receipt time to Section 1 within all future BUFR messages written to output by subroutines writsb(), copymg() or equivalent. The tank receipt time is a local extension to Section 1; however, its inclusion in a message is still fully compliant with the WMO BUFR regulations. This new capability is activated via an initial call to new subroutine strcpt(), specifying the time to be appended to Section 1 within all future BUFR messages written to output. This same information can then be read back from an input BUFR message via a call to new subroutine rtrcpt().
Subroutine numtab() was re-written to directly call subroutine numtbd() instead of duplicating all of the code logic in numtbd().
Subroutine nemtba() was re-written to directly call subroutine nemtbax() instead of duplicating all of the code logic in nemtbax().
Documentation was improved within numerous subroutines throughout BUFRLIB, including the addition of docblocks where none previously existed.
The default BUFR master table version number was changed from "12" to "13" within subroutines cmsgini(), dxmini() and msgini().
A new capability was added to allow BUFRLIB print diagnostics and other runtime messages to be redirected somewhere other than the default FORTRAN logical unit #6 (i.e. standard output). This is enabled within an application program by supplying an in-line version of subroutine errwrt() to override the new default version of this subroutine provided within the BUFRLIB. The default version will continue to write to standard output when included within a compilation.
Subroutines cmsgini(), stndrd() and msgwrt() were modified to remove a logical error which assumed that any message whose data section (Section 4) was compressed was also fully standard. In reality, the use of compression only implies that the data section is fully standard and does not necessarily imply that the data description section (Section 3) is also fully standard. BUFRLIB will now address the standardization of Section 3 solely within subroutine stndrd(), independent of whether or not the data in Section 4 are compressed.
Functions LSTRPC and LSTRPS have been marked as obsolete (for future removal from BUFRLIB). The same functionality can now be obtained via the use of function lstjpb().
Subroutine ufbtab() was modified to fix a bug involving the unpacking of character strings which are identical within each subset of a single compressed BUFR message.

May 28, 2008

Subroutine BORT_EXIT was modified to fix a faulty ANSI-C declaration. This had been silently ignored by the IBM CCS compiler but was a portability issue for other compilers.
Subroutines RDTREE and WRTREE were modified to fix a bug which, on rare occasions, caused a segmentation fault due to overflow of internal arrays. This bug only occurred when working with long character strings (i.e. longer than 8 bytes) while using a non-optimized compilation of BUFRLIB.
Subroutine WRITCA, which had been marked as obsolete within a previous version of BUFRLIB, has now been deleted.
A new subroutine PARSTR was added which works like existing subroutine PARSEQ, except that it allows substrings within a string to be separated by one or more occurrences of any given single character (and not just by one or more blank characters). As such, the existing subroutine PARSEQ has been marked as obsolete (for future removal from BUFRLIB), and many other subroutines throughout BUFRLIB have been modified to now use the new subroutine PARSTR.
Subroutine JSTCHR was modified to add a return argument indicating whether the input string was empty. This allows the subroutine to be used in any context where existing subroutine LJUST was being used, and LJUST has now been marked as obsolete (for future removal from BUFRLIB).
Several new subroutines have been added to enable the capability to read BUFR table information from external ASCII master tables instead of from pre-defined DX dictionary files. This is in preparation for the planned future capability to be able to directly decode a BUFR message according to its internal data description section.
The value BMISS (i.e. the BUFR "missing" value), which was defined as a local data value within many separate subroutines, has now been defined as a global parameter within the "bufrlib.PRM" include file. In addition, a new function IBFMS has been added which safely tests a given value to determine whether or not it is "missing", and several existing subroutines throughout BUFRLIB have been modified to now use this new function.
The determination as to whether the local host machine uses the "big-endian" or "little-endian" byte-ordering scheme is now determined at compile time and integrated into BUFRLIB via the use of conditional compilation statements. This allows BUFRLIB to run much more efficiently since it no longer has to constantly re-check the local byte-ordering scheme at run time.
Subroutine DXDUMP was modified to correct a bug which caused the truncation of output reference values longer than 8 digits.
Several global parameters were increased in "bufrlib.PRM". Specifically, MXCDV (the maximum number of data values per subset in a compressed BUFR message) was increased from 2000 to 3000, and MAXMEM (the maximum number of bytes that can be used to store BUFR messages within internal memory) was increased from 50Mb to 75Mb within the "supersized" BUFRLIB.

February 6, 2007

Several global parameters were increased in "bufrlib.PRM". Specifically, MAXTBA, MAXTBB and MAXTBD (the maximum numbers of internal Table A, B and D entries, respectively) were increased from 60, 250 and 250 to 120, 500 and 500, respectively, and MAXJL (the maximum number of internal jump/link table entries) was increased from 16000 to 20000.
Subroutine CKTABA was modified to allow "FRtttsss" and "FNtttsss" (where ttt is the message type and sss is the message subtype) as valid Table A mnemonics for foreign BUFR messages. Previously, only "NCtttsss" had been allowed.
Subroutines GETS1LOC and IUPBS01 were modified to provide two additional options for unpacking values from Section 1 of a BUFR message. Specifically, "CENT" now unpacks the century and "YCEN" now unpacks the year of the century.
Subroutine PKBS1 was modified to provide several additional options for directly packing values into Section 1 of a BUFR message. Specifically, "YEAR", "MNTH", "DAYS", "HOUR", "CENT" and "YCEN" now pack the message year, month, day, hour, century and year of century, respectively, and "MTYP" and "MSBT" now pack the message type and subtype, respectively.
Subroutine MAXOUT was modified to allow it to be called with a special flag value of "0", indicating that output BUFR messages should be set to the maximum allowable record length. In addition, a sanity check was added to prevent this record length from being set to a value greater than the maximum allowable.
For the printing of flag table values, subroutines UFBDMP and UFDUMP were modified to include an equivalent listing of the bits that were actually set.
Subroutine UFBPOS was modified to remove an unnecessary (and incorrect!) initialization statement. This had been silently ignored by the IBM CCS compiler but was a portability issue for other compilers.
Subroutine UFBTAB was modified to add a required declaration for a local character variable. This had been silently ignored by the IBM CCS compiler but was a portability issue for other compilers.
Subroutine RDUSDX was modified to abort if it encounters a user-defined BUFR message whose message type is set to 11. This value is reserved for internal dictionary messages.

January 31, 2006

Documentation was improved and/or clarified within many existing routines throughout BUFRLIB.
Global parameter MAXMEM (the maximum number of bytes that can be used to store BUFR messages internally) was increased from 16Mb to 50Mb, and global parameter MAXTBA (the maximum number of entries in the internal BUFR Table A) was increased from 50 to 60. In addition, all global parameters were moved into a new global INCLUDE file "bufrlib.PRM", rather than continuing to hardcode the same parameter values in every individual source file where they were needed. This will allow future changes to any of these parameter values to be made much more easily.
An additional CCS compilation of BUFRLIB (libbufr_s_64.a) is now being maintained via the makefile. This new version is identical to the existing libbufr_4_64.a compilation, except that several parameter values within "bufrlib.PRM" are set much larger in order to allow extremely large BUFR messages (i.e. up to 2.5Mb) to be processed.
The capability to compress output BUFR messages has now been directly incorporated into subroutines WRITSB and WRITSA, whereas previously it had been necessary to instead call the separate subroutines WRITCP and WRITCA, respectively. The use of compression can now be easily toggled on or off (with "off" as the default if left unspecified) via new subroutine CMPMSG. As such, subroutine WRITCA has now been marked as obsolete (for future removal from BUFRLIB), since the same functionality can now be obtained by calling the new subroutine CMPMSG, followed by a call to WRITSA. In a similar manner, WRITCP has now been modified to directly call subroutines CMPMSG and WRITSB, although it is being retained as a distinct subroutine within BUFRLIB (rather than also being marked as obsolete) out of consideration for the large number of existing application programs which use it.
A new function IUPBS01 was added which works like existing function IUPBS1, except that it uses a descriptive mnemonic rather than a hardcoded byte number in order to specify the value to be unpacked from Section 0 or Section 1 of a BUFR message. This allows the same function call to work on messages encoded using either BUFR edition 3 or BUFR edition 4 (rather than having to pass in different byte numbers depending on the edition!), and it also allows values encoded across multiple bytes (e.g. section lengths, 4-digit years, etc.) to be easily unpacked as well. As such, the existing function IUPBS1 has been marked as obsolete (for future removal from BUFRLIB), and many other subroutines throughout BUFRLIB (e.g. UPDS3, DATEBF, DUMPBF, STNDRD, CKTABA, NMBYT, MSGWRT, RDBFDX, etc.) have been modified to now use the new function IUPBS01. In addition, a new function IUPVS01 was added which calls IUPBS01 in an in-line fashion, and existing function IUPVS1 (which had similarly called IUPBS1 in an in-line fashion) has now been marked as obsolete.
A new subroutine PKVS01 was added which works like existing subroutine PKVS1, except that it uses a descriptive mnemonic rather than a hardcoded byte number in order to specify the value to be stored into Section 0 or Section 1 of all future output BUFR messages. This allows the same subroutine call to work on messages encoded using either BUFR edition 3 or BUFR edition 4 (rather than having to pass in different byte numbers depending on the edition!), and it also allows values encoded across multiple bytes (e.g. 4-digit years, originating centers and subcenters, etc.) to be easily overwritten as well. As such, the existing subroutine PKVS1 has been marked as obsolete (for future removal from BUFRLIB). In a similar manner, a new subroutine PKBS1 was also added to replace existing subroutine OVRBS1, which has now itself also been marked as obsolete.
A new subroutine CNVED4 was added which, given a BUFR message encoded using BUFR edition 3, creates and outputs an equivalent message encoded using BUFR edition 4. This subroutine can be called by an application program, or it can alternatively be activated in an in-line fashion via a call to new subroutine PKVS01 using the descriptive mnemonic "BEN" (i.e. BUFR edition number) with a corresponding value of "4".
Subroutines NEMTAB, NUMTAB, TABENT and TABSUB were modified to support the Table C operators 2-07-YYY and 2-08-YYY, which are new to BUFR with the advent of edition 4.
Subroutines COPYST, WRITST and STANDARD, which had been marked as obsolete within a previous version of BUFRLIB, have now been deleted.
The default BUFR master table version number was changed from "4" to "12" within subroutines CMSGINI, DXMINI and MSGINI.
A bug was corrected in subroutine STNDRD in order to ensure that byte 4 of Section 4 is always properly zeroed out.
A bug was corrected in subroutine PARUTG which was preventing 1-bit delayed replication factors from being directly read via a call to subroutine UFBINT.
A bug was corrected in subroutine WRCMPS which was causing a character compression array to be improperly initialized. In addition, a local parameter was increased to allow up to 4000 subsets to be written into a single compressed BUFR message.
Subroutine UFBMEM was modified to not abort when there are either too many messages read in or too many bytes read in (i.e., .gt. array limits passed in), but rather to just process the limiting number of messages and/or bytes and print a diagnostic.
Subroutine CLOSMG was modified to override logic that had always written out messages 1 and 2 even when they contained zero subsets (it assumed these contained the dump center and processing time in Section 1). Now, if the unit number argument is passed in as a negative number the first time this routine is called by an application program, ALL empty messages are skipped (i.e., assumes that messages 1 and 2 do not contain dump center and processing time). This remains set for all subsequent calls to CLOSMG for a particular file, regardless of the sign of the unit number (CLOSMG is called by other BUFRLIB routines which always pass in a positive unit number).
A new function IGETDATE was added which unpacks and returns the Section 1 date-time from an input BUFR message, in format of either YYYYMMDDHH or YYMMDDHH depending on the value requested via the most recent call to subroutine DATELEN. The same logic had been repeated within numerous existing subroutines throughout BUFRLIB and has now been consolidated into this single subroutine that can itself be called from wherever it is needed.
A new subroutine GETLENS was added which unpacks and returns the individual section lengths from an input BUFR message. The same logic had been repeated within numerous existing subroutines throughout BUFRLIB and has now been consolidated into this single subroutine that can itself be called from wherever it is needed.
A new subroutine RDMSGW was added which reads the next padded BUFR message from a given BUFR file. The same logic had been repeated within numerous existing subroutines throughout BUFRLIB and has now been consolidated into this single subroutine that can itself be called from wherever it is needed.
A new function PKFTBV was added which computes and returns the value equivalent to the setting of a specified bit within a flag table of a specified width. In addition, a new subroutine UPFTBV was also added which functions as the logical inverse, i.e. given a mnemonic and corresponding flag table value, it computes and returns the equivalent bit settings.
A new subroutine UFBPOS, which allows a user to directly point at and read a specified subset from within a specified message in an input BUFR file, was added to BUFRLIB. Previously, this logic existed as an in-line subroutine within a separate application program.
A new subroutine GETABDB, which returns internal BUFR table information in a pre-defined ASCII format, was added to BUFRLIB. Previously, this logic existed as an in-line subroutine within a separate application program.
Subroutine READMG was modified to be able to handle BUFR messages which are not padded out to an 8-byte boundary and for which it had therefore previously been necessary to instead call the separate subroutine READIBM. Logic was also added to allow the option of having READMG behave like the separate subroutine READFT, so that it will not abort when a read error is encountered but rather will treat it the same as an end-of-file condition. This option is activated by passing in the negative of the usual logical unit number. In summary, READMG can now itself properly read from any FORTRAN-blocked file of BUFR messages, and therefore the existing subroutines READIBM, IREADIBM, READFT and IREADFT have now all been marked as obsolete (for future removal from BUFRLIB).
A set of generic C-language functions for reading/writing BUFR messages from/to generic BUFR files (which may or may not contain FORTRAN-blocking and/or message padding) was added to BUFRLIB. These functions (CCBFL, COBFL, CRBMG, CWBMG and RBYTES) are primarily intended for use by separate application programs (such as cwordsh), but are themselves being directly incorporated into BUFRLIB in order to prevent such application programs from having to directly link to certain COMMON blocks and parameter sizes internal to BUFRLIB.
Function MOVA2I is marked as obsolete (for future removal from BUFRLIB). It is present in the W3 Libraries (in C language) and is no longer called by any BUFR Archive Library routines. A warning message is now printed instructing users to migrate to MOVA2I in the W3 Libraries.
Subroutine UFBTAB was modified to work for compressed BUFR messages. An option to return only the subset count (when the input unit number is less than zero) was also added.
Subroutine COPYSB was modified to now write out a compressed subset/message if the input subset/message is compressed (before this subroutine could only write out an uncompressed subset/message regardless of the compression status of the input subset/message).

December 21, 2004

New subroutines ISTDESC, RESTD, WRDESC, CADN30, STDMSG and STNDRD have been added to provide the capability to expand Section 3 of output BUFR messages until they are completely "standard" according to the WMO FM-94 regulations. The logic is activated via an initial call to STDMSG.
Subroutine XMSGINI has been removed. It had been included in a previous BUFRLIB version as an indirect way of "standardizing" compressed messages, but the same logic is now fully integrated into CMSGINI and is activated via a separate initial call to new subroutine STDMSG (see 1).
Subroutine STANDARD has been marked as obsolete (for future removal from BUFRLIB) in favor of a new subroutine STNDRD which more completely "standardizes" Section 3. The old subroutine (i.e. STANDARD) would always just break down the top-level Table A descriptor by one level, so that, unless this "one level deep" expansion happened to consist of all standard descriptors, the resulting BUFR message was still non-standard. Contrarily, the new logic will recursively break down successive sequence descriptors for as long as needed until all appearing in Section 3 are themselves standard or else, at a minimum, preceded with the 206YYY "bypass" operator (note: this recursive logic is written using C for portability reasons, since not all FORTRAN 77 compilers support recursion!). In addition, STNDRD has other advantages over STANDARD as well; namely, it contains safety checks which prevent overflow of the message array that is passed to it, and it also is more directly integrated into BUFRLIB and can be automatically activated in-line via a separate initial call to new subroutine STDMSG (see 1).
Subroutine WRITSA was modified to fix a bug which, in certain situations, prevented one or more BUFR messages from being returned to the calling program within the memory arrays. In addition, a new subroutine WRITCA was added which functions exactly like WRITSA except that it works on compressed messages.
Subroutines WRCMPS and RDCMPS were modified to fix a bug in the compression algorithm which occurred when all subsets in a single message contained identical character strings. Separate corrections were also made to each of these subroutines to fix a few unrelated minor bugs.
Subroutine UFDUMP was modified to add a fuzziness test for the "missing" value and to add an interactive, scrolling print capability similar to UFBDMP.
Subroutine UFBDMP was modified to automatically use READLC when reading "long" character strings, similar to an existing capability within UFDUMP.
Documentation was improved and/or clarified in many existing subroutines throughout BUFRLIB.
Subroutines COMPRES and READ2C have been removed. The same functionality can be obtained by using subroutine WRITCP.
Subroutines IREADERS, READERS and READTJ have been removed, as they were nothing more than wrappers for READMG and had been marked as obsolete within a previous BUFRLIB version.
Subroutines READERM, IREADERM and IRDERM have been removed. They had been superseded functionally by (the more-portable!) subroutine READIBM and had been marked as obsolete within a previous BUFRLIB version.
Parameter MXMSGL (the maximum number of bytes in a BUFR message) was increased from 20K TO 50K bytes in the following subroutines: BFRINI, CKTABA, CLOSMG, CMSGINI, COPYBF, COPYMG, COPYSB, CPYMEM, CPYUPD, DXMINI, MAXOUT, MESGBC, MESGBF, MINIMG, MSGINI, MSGUPD, MSGWRT, NMBYT, POSAPN, POSAPX, RCSTPL, RDBFDX, RDCMPS, RDMEMM, RDMEMS, RDMGSB, RDTREE, READERME, READFT, READIBM, READLC, READMG, READMM, READSB, REWNBF, SUBUPD, UFBGET, UFBINX, UFBMEM, UFBTAB, UFBTAM, WRCMPS, WRITDX, WRITLC, WRITSA and WRTREE. (Note: this is not included in the Docblock history in these routines.)
Subroutines READERME, READIBM, DATEBF and DUMPBF were modified to make the test for the string 'BUFR' portable to EBCDIC machines.
Subroutine WRTREE was modified to use double-precision arithmetic within an internal statement function, in order to correct for a truncation problem that could occur in the case of very large computed values.
Subroutine COPYST has been marked as obsolete (for future removal from BUFRLIB). The same functionality can be obtained by calling new subroutine STDMSG, followed by a call to COPYMG.
Subroutine WRITST has been marked as obsolete (for future removal from BUFRLIB). The same functionality can be obtained by calling new subroutine STDMSG, followed by a call to CLOSMG.
A new option IO="NODX" has been added to subroutine OPENBF. In this case, the subroutine behaves exactly as if it had been called with IO="OUT", except that DX dictionary messages are not written out to logical unit LUNIT.
Subroutine WRDLEN was modified to keep track of whether it has already been called by one of the other BUFRLIB subroutines and, if so, to then immediately return (without proceeding any further) every time it is subsequently called.
Subroutines OPENBF, UFBINT, UFBOVR, UFBREP, UFBSEQ, UFBSTP and WRDLEN were all modified to fix similar portability bugs whereby the values of some internal variable(s) which keep track of whether the subroutine has already been called were not being explicitly preserved with a SAVE statement.
New subroutine PKVS1 was added which calls OVRBS1 in an in-line fashion and therefore allows easier overwriting of default values in Section 1 of output BUFR messages. The new methodology can also overwrite the value of byte 8 in Section 0 (i.e. BUFR edition number) if desired.
New function IUPVS1 was added which calls IUPBS1 in an in-line fashion and therefore allows easy unpacking of Section 1 values from BUFR messages that have already been read into the internal memory arrays by subroutine READMG or equivalent. The new methodology can also unpack the value of byte 8 in Section 0 (i.e. BUFR edition number) if desired.
Subroutine ADDATE was modified to fix a bug in calculating the number of days in February for years which are multiples of 100 but not of 400.
Subroutine MESGBC was modified to allow the option of operating on a BUFR message that has already been read into the internal memory arrays by subroutine READMG or equivalent.
New subroutine DXDUMP was added which outputs an ASCII-formatted copy of the information embedded within the DX dictionary messages of a BUFR file. It is especially useful for learning the contents of archived BUFR files, and the output is in a format suitable for subsequent input to OPENBF as a user-defined dictionary tables file.
Subroutines DATELEN, DATEBF and DUMPBF were all modified to call subroutine WRDLEN to initialize local machine information (in case it has not already been called). These routines do not require this information but they may now or someday call other routines that do require it.

November 4, 2003

This is the first "unified" BUFR Archive Library including components from the regular NCEP production machine version (whose implementation history is documented to this point), the decoder version (previously on a workstation but now on the IBM Frost and Snow machines), and a checkout NCEP/EMC grid-to-obs verification version. This version is portable to all platforms (as necessary for WRF), contains docblocks for each routine with a complete program history log, and outputs more complete diagnostic information when routines terminate abnormally, unusual things happen or for informational purposes.

The following libraries are now generated on the NCEP IBM Frost and Snow machines:

libbufr_4_64.a – 4-byte reals, 4-byte integers, 64-bit executable compilation
libbufr_8_64.a – 8-byte reals, 8-byte integers, 64-bit executable compilation
libbufr_d_64.a – 8-byte reals, 4-byte integers, 64-bit executable compilation
libbufr_4_32.a – 4-byte reals, 4-byte integers, 32-bit executable compilation

The first three are compiled exactly the same as the three libraries noted in the previous 05-27-2003 implementation (they are just renamed). The fourth library is compiled identically to the previous decoder-specific version on Frost and Snow (libdecod_bufr_32.a). It is compiled with optimization level 3 (-O3) for both Fortran and c routines and will be linked into the production decoder programs in place of libdecod_bufr_32.a.

The following routines have been added to the BUFR archive library:

Subroutine BORT2 which prints (to STDOUT) two given error strings and then calls BORT_EXIT (see 7 below) to abort the application program calling the BUFR Archive Library software. It is similar to existing subroutine BORT, except BORT prints only one error string.
Function IUPBS1 which, given a BUFR message contained within array MBAY, unpacks and returns the binary integer contained within byte NBYT of Section 1 of the BUFR message. This was present in the original decoder- specific version of the library.
Subroutine OVRBS1 which, given a BUFR message contained within array MBAY, packs and stores the value of a binary integer into byte NBYT of Section 1 of the BUFR message, overwriting the value previously stored there.
Subroutine UPDS3 which, given a BUFR message contained within array MBAY, unpacks and returns the descriptors contained within Section 3 of the BUFR message. This was present in the original decoder-specific version of the library.
New library function MOVA2I replaces the W3LIB c-version previously called by DATEBF and DUMPBF. It is now called by new subroutines MESGBC (see 6) and REWNBF (see 11) as well. This change removes any dependency upon the W3LIB, since no other BUFR Archive Library routines call any W3LIB routines. It was converted to Fortran 77 because the c-version does not work properly when compiled with 32-bit executable compilation and linked into a Fortran source copiled with 8-byte real and integer word length.
Subroutine MESGBC reads past any BUFR table (dictionary) or dummy (center or dump time in dump files) messages in a BUFR file (if there are any) and returns the message type for the first report data message found. It also determines whether or not this first report data message is compressed BUFR. This subroutine is identical to MESGBF except MESGBF only reads past dictionary messages and MESGBF does not return any information about compression.
C subroutine BORT_EXIT terminates the application program calling the BUFR software and returns an implementation-defined non-zero status code to the executing shell script. (See 1 and 26.)
Subroutine RDMGSB opens a BUFR file in logical unit LUNIT for input operations, then reads a particular subset into internal subset arrays from a particular BUFR message in a message buffer. This is based on the subset number in the message and the message number in the BUFR file. This was present in the original verification-specific version of the library.
Subroutine SUBUPD packs up the current subset within memory and then tries to add it to the BUFR message that is currently open within memory for LUNIT. If the subset will not fit into the currently open message, then that message is flushed to LUNIT and a new one is created in order to hold the current subset. If the subset is larger than an empty message, the subset is discarded and a diagnostic is printed. This subroutine is identical to existing BUFR Archive Library subroutine MSGUPD except SUBUPD does NOT pad the packed subset. This was present in the original verification-specific version of the library.
Subroutine UFBINX either opens a BUFR file for input operations (if it is not already opened as such), or saves its position and rewinds it to the first data message (if BUFR file already opened), then (via a call to BUFR Archive Library subroutine UFBINT) reads specified values from internal subset arrays associated with a particular subset from a particular BUFR message in a message buffer. The particular subset and BUFR message are based based on the subset number in the message and the message number in the BUFR file. Finally, this subroutine either closes the BUFR file (if is was opened here) or restores it to its previous read/write status and position (if is was not opened here). This was present in the original verification-specific version of the library.
Subroutine REWNBF which will either: 1) store the current parameters associated with a BUFR file (read/write pointers, etc.), set the file status to read, then rewind the BUFR file and position it such that the next BUFR message read will be the first message in the file containing actual subsets with data; or 2) restore the BUFR file to the parameters it had prior to 1) using the information saved in 1). This allows information to be extracted from a particular subset in a BUFR file which is in the midst of being read from or written to by an application program. This was present in the original verification-specific version of the library.
Subroutine UFBIN3 reads specified values from the current BUFR data subset where the data values correspond to mnemonics which are part of a multiple- replication "level" sequence within another multiple-replication "event stack" sequence. This subroutine is designed to read event information from "PREPFITS" type BUFR files (currently the only application which reads PREPFITS files is the verification program GRIDTOBS, where UFBIN3 was previously an in-line subroutine). The existing analogous subroutine UFBEVN should be used to read information from "PREPBUFR" type BUFR files. This was present in the original verification-specific version of the library.
Function NEVN accumulates all data events for a particular data value and level and returns them to the calling program. The value of the function itself is the total number of events found. {This function should only be called by UFBIN3 (see 12), which, itself, is called only by verification application program GRIDTOBS, where it was previously an in-line subroutine. In general, NEVN does not work properly in other application programs at this time.} This was present in the original verification-specific version of the library.
Subroutine READLC returns a character data element associated with a particular subset mnemonic from the internal message buffer. It is designed to be used to return character elements greater than the usual length of 8 bytes. It currently will not work for compressed BUFR mesaages.
Subroutine WRITLC packs a character data element associated with a particular subset mnemonic from the internal message buffer. It is designed to be used to store character elements greater than the usual length of 8 bytes.
Subroutine WRITST generates a standardized version of the current BUFR message in internal memory and writes it to the output file (not sure if it works properly).
Subroutine COPYST generates a standardized version of the current BUFR message read using READMG and writes it intact as a record to the output file.
Subroutine COMPRES compresses subsets in BUFR messages previously read using BUFR Archive Library subroutine READMG or equivalent. It then generates a new bufr message consisting of the compressed subsets. Note that subsets in the output compressed message may have been read from different (adjacent) input messages. Currently the only application program which calls this subroutine is BUFR_COMPRESS, where COMPRES was previously an in-line subroutine).

The following routines in the BUFR archive library have been modified:

Subroutine READ2C reads a subset into compression maxtrix arrays in preparation for generating compressed BUFR messages. This had been an in- line subroutine in the application program BUFR_COMPRESS and is currently called only by BUFR Archive Library subroutine COMPRES (see 18).
Subroutine UPBB modified to make certain zero is returned for zero bits input and to make logic consistent with logic in UPB. (See also 30 for UPBB.)
Subroutine UFBTAB modified to not abort when there are too many subsets coming in (i.e., .gt. array limit passed in), but rather to just process the limiting number of reports and print a diagnostic. It is also modified to call subroutine REWNBF when the BUFR file is already opened (this is taken from the verification version of UFBTAB and allows specific subset information to be read from a file in the midst of being read from or written to), before OPENBF was always called and this would have led to an abort of the application program. (See also 29 for UFBTAB.)
Subroutine CKTABA modified to not abort when the Section 1 message subtype does not agree with the Section 1 message subtype in the dictionary IF the message type mnemonic is not of the form "NCtttsss", where ttt is the BUFR type and sss is the BUFR subtype. This allows program PREPOBS_PREPDATA to specify different message subtypes for the same message type. (See also 27 and 43.)
Subroutine OPENBF modified to accept 'NUL' as the second (I/O) argument. IO='NUL' prevents the BUFR Archive Library software from actually trying to access or write to the BUFR file (designed only for use with library subroutine WRITSA). This was present in the original decoder-specific version of the library.
Subroutine CLOSBF modified to not close the BUFR file if it was opened as 'NUL' by OPENBF (see 23). This was present in the original decoder-specific version of the library.
Subroutine MSGWRT modified to not write to the BUFR file if it was opened as 'NUL' by OPENBF (see 23). This was present in the original decoder- specific version of the library.
Subroutine BORT modified to call new BUFR Archive Library subroutine BORT_EXIT (see 7 above) rather than c function EXIT with argument 49 {"CALL EXIT(49)"}. Since EXIT is an intrinsic c function, it expects arguments to be passed by value rather than by reference as in done in Fortran. This has caused an unpredictable status code to be passed back to the executing shell script, in some cases even ZERO!! This change will ensure an non-zero status is always returned.
Suboutines CKTABA, DATEBF, DUMPBF and function I4DY modified such that date calculations no longer use floating point arithmetic. This can lead to round off error and an improper resulting date on some machines (e.g., NCEP IBM Frost/Snow). This change increases portability of the BUFR Archive Library. (See also 22 and 43 for CKTABA.)
Parameter MAXMSG (the maximum number of BUFR messages which can be stored internally) increased from 50000 to 200000 in the following subroutines: CPYMEM, RDMEMM, RDMEMS, READMM, UFBMEM, UFBMMS, UFBMNS, UFBRMS and UFBTAM. This may be necessary in the future for BUFR files with many, many messages.
Parameter MAXJL (the maximum number of Jump/Link table entries) increased from 15000 to 16000 in the following routines: BFRINI, CONWIN, COPYMG, CPYMEM, DRFINI, DRSTPL, GETWIN, INCTAB, INVCON, INVMRG, INVTAG, INVWIN, LSTJPB, LSTRPC, LSTRPS, MAKESTAB, MSGINI, NEWWIN, NVNWIN, NWORDS, NXTWIN, PARUTG, RCSTPL, RDCMPS, RDTREE, READNS, TABENT, TABSUB, TRYBUMP, UFBCPY, UFBCUP, UFBDMP, UFBEVN, UFBGET, UFBINT, UFBOVR, UFBREP, UFBRP, UFBRW, UFBSEQ, UFBSP, UFBSTP, UFBTAB, UFBTAM, UFDUMP, USRTPL, WRCMPS and WRTREE. This was present in the original verification-specific version of the library.
The following routines are modified to make the BUFR Archive Library big-endian/little-endian independent: IPKM, IUPM, PKB, PKC, UPB and UPBB. This was present in the original decoder-specific version of the library and increases the portability of the BUFR Archive Library.
Subroutine BFRINI modified to initialize variable JSR as ZERO in new COMMON block /BUFRSR/. This was present in the original verification-specific version of the library. (See also 29 for BFRINI.)
Subroutine RCSTPL modified to increase the maximum number of levels of recursion (parameter MAXRCR) from 50 to 100. This was present in the original verification-specific version of the library. (See also 29 and 43 for RCSTPL.)
Subroutine WRCMPS modified to save logical variables WRIT1 and FLUSH in global memory. This fixed a bug in this subroutine which could lead to messages being written out before being full. (See also 29 for WRCMPS.)
Subroutine RDTREE modified to fix a bug which could only occur when the last element in a subset is a character. (See also 29 for RDTREE.)
Subroutine UFDUMP modified to handle print of character values greater than 8 bytes. (See also 29 for UFDUMP.)
Subroutine UFBEVN modified to save the maximum number of events found for all data values specified amongst all levels returned as variable MAXEVN in new COMMON block /UFBN3C/ and to add call to BORT if BUFR file is open for output. (See also 29 for UFBEVN.)
Subroutine NEMOCK modified to expand non-zero return into -1 for length not 1-8 characters and -2 for invalid characters (return only -1 before for all problematic cases)
Subroutine NUMBCK modified to expand non-zero return into -1 for invalid character in position 1, -2 for invalid characters in positions 2 through 6, -3 for invalid characters in positions 2 and 3 due to being out of range, and -4 for invalid characters in positions 4 through 6 due to being out of range (return only -1 before for all probelmatic cases)
Subroutine WTSTAT modified to correct a "typo" in test for valid value for "IM".
Subroutines ELEMDX, PARSEQ, PARUSR, PARUTG, PKC, RDUSDX, SEQSDX, STRING, UFBINT, UFBOVR, UFBREP, UFBSTP and VALX modified to call new BUFR Archive Library subroutine BORT2 (see 1).
Subroutine MAKESTAB modified to allow for the possibility that a connected file may not contain any dictionary table info (e.g., an empty file). Subsequent connected files which are not empty will no longer get tripped up by this. (This change avoids the need for an application program to disconnect any empty files via a call to CLOSBF.) (See also 29 for MAKESTAB.)
Subroutine READTJ modified to simply call BUFR Archive Library subroutine READMG rather than being a clone of it. At one time it performed different functions than READMG, but that has not been the case since the 2000-09-19 BUFR Archive Library implementation.
Subroutines CKTABA, CMSGINI, NUMTAB, PARUSR, PARUTG, RCSTPL, USRTPL, WRDLEN, WRTREE and XMSGINI modified to correct some minor bugs (uninitialized variables, etc.) (see subroutine DOCBLOCKS for more information). (See also 29 for PARUTG, RCSTPL, USRTPL, WRTREE and 32 for RCSTPL.)
Subroutine UFBDMP modified to add "fuzziness" about 10E10 in test for a missing value (rather than true equality as before) because some missing values (e.g., character strings < 8 characters) were not getting stamped out as "MISSING". Also added option to print values using format edit descriptor "F15.6" if input argument LUNIN is < zero. If LUNIN is > zero edit descriptor expanded from "G10.3" to "G15.6". (See also 29 for UFBDMP.)

May 19, 2003

The following changes have been made in the BUFR Archive Library:

Subroutine CLOSMG - to correct a problem introduced in the previous (May 2002) implementation which prevented the dump center time and initiation time messages from being written out (affected program BUFR_DUMPMD, if it were recompiled, in the data dumping process)
Subroutine UFBREP - to work properly for descriptors tied to a pivot descriptor in delayed replicated sequences (involved disabling the parsing switch which controlled checking for presence in the same replication group - UFBREP does not need this check, and it interferes with what UFBREP can do otherwise)
Subroutine UFBSEQ - to fix cases where delayed replication is at end of subset, or when a requested sequence is missing; also corrected the logic array of exit conditions for the subroutine, previously, in some cases, proper exits were missed, generating bogus error messages, because of several miscellaneous bugs which are now removed
Subroutine UPB - to make certain zero is returned for zero bits input
The following subroutines are modified to replace calls to Fortran Insrinsic Function ICHAR with calls to NCEP W3LIB c-function MOVA2I: DATEBF and DUMPBF. This change increases portability of the BUFR Archive Library because MOVA2I copies a bit string from a Character*1 variable to an integer variable. It is intended to replace the Fortran Intrinsic Function ICHAR, which only supports 0 <= ICHAR(A) <= 127 on the IBM SP. If "A" is greater than 127 in the collating sequence, ICHAR(A) does not return the expected bit value. This function can be used for all values of ICHAR(A) between 0 and 255. This change increases portability of the BUFR Archive Library and is, in fact, necessary on the NCEP IBM Frost and Snow machines.

The BUFR Archive Library on Frost and Snow is compiled using optimization level 4 (-O4) for Fortran routines and optimization level 3 (-O3) for c routines. The previous BUFR Archive Library on the IBM-SP's had used -O3 for the default filenames and -O4 for a second set of filenames with the string "_O4" appended to the end.

The following libraries are generated on the NCEP IBM Frost and Snow machines:

libbufr_4.a – 4-byte reals, 4-byte integers, 64-bit executable compilation
libbufr_8.a – 8-byte reals, 8-byte integers, 64-bit executable compilation
libbufr_d.a – 8-byte reals, 4-byte integers, 64-bit executable compilation

May 14, 2002

A number of routines in the BUFR Archive Library have been modified. These changes include:

Entries IREADMM, IREADNS, IREADSB, IREADERS, IREADIBM, IREADFT and ICOPYSB changed to functions. Entries MRGINV, MINIMG, DATELEN, NENUBD, NENUAA, JSTNUM, JSTCHR and READERS changed to subroutines (note that READERS now simply calls READSB since it was an entry point at the top of READSB and was thus already an alias to it). Converted all entry points to subroutines or functions in order to increase portability to other platforms (e.g., the NESDIS CEMSCS machine).
Entries DXMINA, DXMINB, DXMIND and SUBUPD removed because they are obsolete.
Added XMSGINI for capacity to expand section three. XMSGINI has the capacity to write a fully expanded section three descriptor set into BUFR messages. Created specifically for NESDIS so they can send files out without local sequence descriptors. This "capacity" is not fully functional, it is currently activated by changing WRCMPS.to call it rather than CMSGINI, which writes sections 0,1,2,3 for compressed messages in the usual way. XMSGINI is included because it is useful for particular situations as is (aka NESDIS), and at some point could be integrated as a more direct form of STANDARDizing messages for export or whatever.
Included in-line compression function (subr. CMSGINI, WRITCP, WRCMPS added).
Improved RDCMPS and UFBSEQ for generality. Previously RDCMPS and UFBSEQ would not recognise compressed delayed replication as a legitimate data structure.
Removed old CRAY compiler directives in: COPYSB, CPYUPD, DRSTPL, GETWIN, INVMRG, MVB, NEWWIN, NXTWIN, RCSTPL, READSB, UFBDMP, UFBGET, UFBINT, UFBOVR, UFBRW, UFBTAB, UFBTAM and USRTPL.
Added new subroutine UFDUMP which is like UFBDMP, but prints subset element contents in more detail, omitting the pointers, counters, and other more esoteric information describing the internal subset structures. Each subroutine, UFBDMP and UFDUMP, is useful for different diagnostic purposes, but in general UFDUMP is more useful for just looking at the data elements.
Corrected error in READSB relating to certain foreign filetypes.
Added new subroutine DRFINI which initializes delayed replication factors, and allocates the space in the full word buffer for their contents explicitly. This is done implictly by UFBINT in a more limited way. DRFINI enables, for instance, the subsequent use of UFBSEQ to write data directly into delayed replicated sequences.
Added new subroutine MAXOUT which allows users to control the record length of output BUFR messages created.
Added new subroutine NUMTBD which is used by XMSGINI, in expanding the section 3 descriptor list.
Added new subroutine CAPIT which capitalizes a string of characters. This enables the use of mixed case in the unit section of the ASCII BUFR tables. An example; a program which generates an ASCII BUFR table from the "Master Table B", might end up copying some units fields in mixed or lower case. If the units are 'Code table' or 'Flag table' or certain other unit designations, the table will be parsed incorrectly, and the data read or written incorrectly as a result. This makes sure all unit designations are seen by the parser in upper case to avoid these types of problems.
Removed subroutine JSTIFY because it was a dummy subroutine with two entry points for left justifying two different types of character strings. Part of conversion of entry points to separate subroutines or functions. See number 1 above.
Removed subroutine NENUCK because it was a dummy subroutine with two entry points for checking the BUFR mnemonic table. Part of conversion of entry points to separate subroutines or functions. See number 1 above.

August 15, 2001

Parameter MAXMEM (the maximum number of bytes required to store all messages internally) was increased from 8 MBYTES TO 16 MBYTES in the following subroutines: CPYMEM, RDMEMM, RDMEMS, READMM, UFBMEM, UFBMMS, UFBMNS, UFBRMS and UFBTAM.
Subroutine UFBTAM modified to not abort when there are too many subsets coming in (i.e., .gt. array limit passed in), but rather to just process the limiting number of reports and print a diagnostic.

September 19, 2000

A number of routines in the BUFR Archive Library have been modified. These changes include:

Consolidated logic that had been replicated in message decoding subroutines READMG, READFT, READTJ, READERM, READERME, RDMEMM and READIBM into a single new subroutine CKTABA (called by these subroutines). On top of this CKTABA now recognizes a variety of Section 3 formats, including compression indicators and "standard" BUFR. Thus, compressed and standard BUFR messages can now be read in via these message decoding subroutines.
The subset decoding subroutine READSB now calls a new subroutine RDCMPS which allows it to read subsets from compressed BUFR messages.
Subroutine RDTRER has been removed. It had been called by READERS to decode ERS scatterometer data from compressed BUFR messages. The change in READSB (see 2) allows READERS to be changed from a subroutine to an entry point at the top of READSB since it is now essentially an alias to READSB.
Subroutine UNCMPS has been removed. It had been called by READERM, READERME and READIBM to uncompress BUFR messages in foreign (i.e., standard ) BUFR files (e.g., ERS scatterometer data). This is a result of change 1 above.
Added capability to encode and decode data using the operator descriptors (BUFR Table C) for changing width and changing scale. Subroutines modified were: NEMTAB, NEMTBD, NUMTAB and TABSUB
Enlarged arrays to allow processing messages up to 20000 bytes. Routines modified were: BFRINI, CLOSMG, COPYBF, COPYMG, COPYSB, CPYMEM, CPYUPD, IRDERM, MESGBF, MINIMG, MSGINI, MSGUPD, MSGWRT, POSAPN, POSAPX, RCSTPL, RDBFDX, RDMEMM, RDMEMS, RDTREE, READERM, READERME, READFT, READIBM, READMG, READMM, READSB, READTJ, UFBGET, UFBMEM, UFBTAB, UFBTAM, WRITDX, WRITSA and WRTREE.
Added subroutine UFBSEQ, like UFBINT except processes specific sequences instead of specific elements.
Added function NMBYT, returns the number of bytes in a message opened for input.

The BUFR Archive Library is now compiled using both optimization level 3 (-O3) and optimization level 4 (-O4). The previous BUFR Archive Library had used only -O4. The -O3 compilation here generates the same archive library names as before. Thus, any code that is recompiled from an unchanged makefile will now link in the appropriate -O3 library, rather than the -O4 library as before. The new -O4 libraries all have the string "_O4" appended to the end of the filename.

Any program that must link to the -O4 BUFR Archive Library when compiled will have to modify its makefile.

July 13, 1999

A number of routines in the BUFR Archive Library have been modified to increase the number of BUFR files which can be opened at one time from 10 to 32. This is necessary in order to process multiple BUFR files under the MPI. The following routines were modified: BFRINI, CHEKSTAB, CLOSMG, CONWIN, COPYMG, COPYSB, CPBFDX, CPYMEM, CPYUPD, DXINIT, ELEMDX, GETWIN, IFBGET, INVCON, INVMRG, INVTAG, INVWIN, LSTJPD, LSTRPC, LSTRPS, MAKESTAB, MSGINI, MSGUPD, NEMTAB, NEMTBA, NEMTBD, NENUCK, NEWWIN, NMSUB, NUMTAB, NVNWIN, NWORDS, NXTWIN, OPENBF, OPENMB, OPENMG, PARUTG, PKTDD, RCSTPL, RDBFDX, RDMEMM, RDMEMS, RDTREE, RDTRER, RDUSDX, READERM, READERME, READERS, READFT, READMG, READNS, READSB, READTJ, STATUS, STRING, TRYBUMP, UFBCNT, UFBCPY, UFBCUP, UFBDMP, UFBEVN, UFBGET, UFBINT, UFBOVR, UFBREP, UFBRP, UFBRW, UFBTAB, UFBTAM, UNCMPS, UPTDD, USRTPL, WRITDX, WRTREE, WTSTAT.
Subroutines READFT, READMG, and READTJ have been modified with semantic adjustments to ameliorate compiler complaints from LINUX boxes.
Added the new subroutine READIBM in order to process "foreign" (non-NCEP) BUFR files which may not be padded. Unlike the subroutine READERM, which performs a similar fuction, READIBM works properly on all platforms and should replace calls to READERM in application programs. (READERM does not work properly on the NCEP IBM-SP machine.)
Added the new subroutine NEMTBAX. It is like subroutine NEMTBA except if the requested mnemonic is not found, it returns rather than calls BORT. This is necessary to support the logic in the new BUFR Archive Library subroutine READIBM (see 3).
Added the new subroutine READMM. It is like subroutine RDMEMM except it advances the value of the message (record) number by one prior to returning to the calling program. This adds another option for application programs which read BUFR files in random access mode (e.g., PREPOBS_OIQCBUFR).
Function IREADMG has been modified to contain two new function entries called IREADMM and IREADIBM. The IREADIBM function calls the new library subroutine READIBM (see 3) and the IREADMM function calls the new library subroutine READMM (see 5).
RDTRER, READERM, READERME and UNCMPS have been modified to expand the maximum number of possible descriptors in a subset from 1000 to 3000.
The maximum number of bytes required to store all messages internally was increased from 4 Mbytes to 8 Mbytes in the following subroutines: RDMEMM, UFBMEM, UFBMMS, UFBMNS and UFBRMS.
The function formerly called VAL$ has been renamed to VALX to remove the possibility of the "$" symbol causing problems on other platforms. In turn subroutine NEMTBB has been modified to call function VALX rather than VAL$.
New subroutines UFBSTP and UFBSP added (UFBSP is called by UFBSTP).

December 14, 1998

Subroutine MSGUPD was updated to bybass the processing of reports that are longer than the length of a BUFR message. Prior to this change, the BUFR Archive Library would issue an abort in the event of this rare, but possible occurrence which occurred at 12Z on 4 December in the RGL suite.
In addition, function I4DY was modified to use 20 as the 2-digit year for windowing to a 4-digit year (00-20 ==> add 2000; 21-99 ==> add 1900). This windowing technique was inadvertently changed to 10 in the previous implementation of the BUFR Archive Library.

November 24, 1998

Function I4DY and subroutine MSGWRT were changed as a result of final Y2K testing of the decoder/ingest system.
I4DY was changed to conform to the NCEP 2-digit year time-window of 1921-2020.
MSGWRT was changed to zero out the padding bytes written at the end of Section 4.

October 27, 1998

The BUFR Archive Library is being modified to correct problems caused by in-lining code with fpp directives. The following subroutines are being changed: DATEBF, MVB, RCSTPL, RDMEMS, RDTREE, RDTRER, UFBGET, UFBRW, UFBTAB, UFBTAM and UPBB.

August 31, 1998

BUFR Archive Library subroutine DATEBF, which returns the center date-time for a BUFR data dump file, is being modified to correct an error which lead to the year being returned in the second argument as 2-digit year when a 4-digit year was requested via a prior call to subroutine DATELEN. The center date returned in the sixth argument, in the form YYYYMMDDHH, was correct in the previous version of this subroutine.

July 8, 1998

The new version of the BUFR Archive Library is Y2K compliant, with additional changes to support expanded machine independence of the code, and to refine, correct, or improve some of the routines within. Although nearly every one of library routines has some change made (mainly because of the introduction of a more general error exit subroutine), the changes largely fall into the first two categories. Three new routines were also added to the BUFR Archive Library for micellaneous puposes.

Y2K Compliance

Y2K compliance in the BUFR Archive Library is downwardly compatible. That is, the new library will read non-Y2K BUFR files as the old one does. However, all two digit years read are represented internally as four digit years, and any files written with the new library will be in Y2K format. A functional conversion of two digit year inputs assumes the years 21-99 are in the twentieth century, while years numbered 00-20 are in the twenty-first. A Y2K BUFR file is identified by a non-zero value in the 18th byte of the message section one, the century byte. At this point users of the library have access to the full four digit year values read by including a signal in their programs via a new entry point called DATELEN. The plan is to have the default set to return two digits of the year during a transition period. This allows implementation of the new BUFR Archive Library into a non-Y2K compliant environment. The susbsequent list of subroutines have been changed for Y2K compliance: BFRINI, DATEBF, DUMPBF, MSGINI, OPENMB, OPENMG, RDMEMM, READERME, READFT, READMG, READTJ.

Machine Independence

Since the last implementation of the BUFR Archive Library, several areas in the code have been identified which are problematic in some way with regards to compiling the library on some computers. Upgrades have been made to the following list of subroutines to address these: CONWIN, INVCON, PARUSR, RDTRER, READERM, READERME, TRYBUMP, UFBEVN, UFBGET, UFBINT, UFBRP, UFBRW, UFBTAB and UNCMPS.

Refinements, Corrections, and Improvements

This is a list of BUFR Archive Library routines which were either in error, or in need of some improvement: IRDERM, NEMTBB, NENUCK, RDBFDX, RDUSDX, STRCLN, STRING, TABENT, UNCMPS and WRTREE.

New Error Exit Subroutine

Many of the BUFR Archive Library routines perform internal testing during operation in order to prevent certain situation from generating mysterious aborts, or, even worse, giving the wrong answers. The original library utilized the Cray library routine ABORT to terminate a program when such a situation was found. The new library uses a new inernal subroutine, BORT, to accomplish this. The list of routines changed for this purpose is as follows: ADN30, CHEKSTAB, CLOSMG, COPYBF, COPYMG, COPYSB, CPYMEM, CPYUPD, DATEBF, DRSTPL, DUMPBF, DXMINI, ELEMDX, GETWIN, IDN30, IFBGET, INCTAB, INVMRG, IPKM, IUPM, JSTIFY, LSTJPB, LSTRPC, LSTRPS, MAKESTAB, MSGINI, MSGUPD, MSGWRT, MVB, NEMTBA, NEMTBB, NEMTBD, NENUCK, NEWWIN, NMSUB, NVNWIN, NXTWIN, OPENMB, OPENMG, OPENBF, PAD, PARSEQ, PARUSR, PARUTG, PKC, POSAPN, POSAPX, RCSTPL, RDBFDX, RDMEMM, RDMEMS, RDUSDX, READDX, READERM, READERME, READERS, READFT, READMG, READNS, READSB, READTJ, SEQSDX, STANDARD, STATUS, STRING, TABENT, TABSUB, UFBCNT, UFBCPY, UFBCUP, UFBDMP, UFBEVN, UFBGET, UFBINT, UFBMEM, UFBMMS, UFBMNS, UFBOVR, UFBQCD, UFBQCP, UFBREP, UFBRMS, UFBTAM, UPTDD, USRTPL, VAL$, WRDLEN, WRITDX, WRITSA, WRITSB, WTSTAT

New Code Added

I4DY the two/four digit year conversion function LJUST a character left justify function OPENBT A dummy entry point which is relevant to users of the READTJ subroutine

April 2, 1998

BUFR Archive Library subroutine STRCLN, which initializes the mnemonic string cache in the BUFR interface, is being modified to enlarge the cache from 50 elements to 1000, maximum.
BUFR Archive Library subroutine STRING manages the mnemonic string cache in the BUFR interface. The mnemonic string cache is a performance enhancing device which saves time when the same mnemonic strings are encountered in a user program, over and over again (the typical scenario). It is being modified to operate a bigger cache, and some optimization of the cache search algorithm is being made in support of a bigger cache.

September 3, 1997

Changes are being made to the BUFR Archive Library to recompile all routines without the -ez compiler option. The removal of this debugging option should speed up the execution of all modules which are linked with BUFR Archive Library routines.
In addition, a new subroutine, STANDARD, is being added to the library. This subroutine "standardizes" NCEP BUFR messages for transmission. It was requested to process hurricane location data.

July 29, 1997

Three BUFR Archive Library subroutines were modified to update the current BUFR version information written into Section 0 of each message: DXMINI, MSGINI and MSGWRT. Version 3 replaces version 2.
Three additional subroutines were modified to enable them to process GOES soundings from NESDIS: IRDERM, RDTRER and READERME.

December 17, 1996

The BUFR Archive Library was modified to make the following changes:

DUMPBF - Corrected error in dump date reader.
RDUSDX - Fixed for some MVS compiler's treatment of internal reads.
RDBFDX - Fixed for some MVS compiler's treatment of internal reads.
UFBINT - Modified to always initialize "USR" array to missing (10E10) when BUFR file is being read.

December 11, 1996

The following subroutines were modified in the BUFR Archive Library:

STATUS - Fixed a long standing bug which occurs in unusual situations. Very low impact.
UFBINT - Removed a hard abort for users who try to write non-existing mnemonics.
UFBRW - Removed a hard abort for users who try to write non-existing mnemonics.
ADDATE - New date arithmetic subroutine added to the library.
DUMPBF - New dump date reader added to the library.
MSGINI - Modified to allow inclusion of minutes in writing the message date into a BUFR message.
READTJ - Specific database ingest message reader added to the library which can attach different BUFR tables if the message type is not recognized in the current ones. Works with a user subroutine called OPENBT which specifies the location(s) of different tables.

November 25, 1996

Several routines in the BUFR Archive Library are being modified to provide more machine independence. The data merging routine is being modified for radiosonde call signs, a return code is being added to UFBINT when mnemonics are not found, and READMG is being modified to exit gracefully when the file is positioned after an end of file

October 9, 1996

The BUFR Archive Library was modified to include 9 additional routines to process ERS scatterometer data (IREADERS, RDTRER, READERS, UNCMPS), perform fault tolerant reading (IREADFT, READFT), and support report part merging (INVMRG, MRGINV, NWORDS).

September 9, 1996

The BUFR Archive Library was separated into 121 BUFR interface routines, which include upgrades and devices for operating the BUFR database.

June 28, 1995

The BUFR Archive Library was modified to increase the size of internal arrays in order to handle bigger files. Coding was also added in order to process ERS scatterometer data which is input from compressed BUFR messages (new subroutine READERME).

January 10, 1995

The BUFR Archive Library was modified slightly to allow for changes in the AVN and FNL PREPBUFR and Q.C. Processing codes (PREPDATA, CQCBUFR, OIQCBUFR, SSIANL).

Original Implementation of BUFR Archive Library - January 6, 1994

Implemented on Cray-YMP as a single monolithic source bufr.f. Only the AVN and FNL PREPBUFR processing and q.c. codes used the BUFR Archive Library initially. These were: PREPDATA, SYNDATA, CQCBUFR, OIQCBUFR, and SSIANL. These had actually been implemented with a non-production version of the library in Jack Woollen's directory September 21, 1993.