BUFR 19.1.3 (MetDB) doesn't build with buildpack

[Ian Culverwell]

... because it looks for TABLEB/D in bufr-19.1.3/extra/TABLED rather than $BUFR_LIBRARY, as it should.

Thus we get error messages like

forrtl: severe (24): end-of-file during read, unit 81, file /data/local/idculv/ROPP/ROPP_tarballs/ropp-7.0/bufr-19.1.3/extra/TABLED
Image              PC                Routine            Line        Source             
testtables         0000000000494ACA  Unknown               Unknown  Unknown
testtables         0000000000493645  Unknown               Unknown  Unknown
testtables         0000000000451A56  Unknown               Unknown  Unknown
testtables         0000000000411275  Unknown               Unknown  Unknown
testtables         0000000000410A31  Unknown               Unknown  Unknown
testtables         0000000000430563  Unknown               Unknown  Unknown
testtables         0000000000408F17  Unknown               Unknown  Unknown
testtables         0000000000404263  Unknown               Unknown  Unknown
testtables         0000000000402B9C  Unknown               Unknown  Unknown
libc.so.6          000000309181ECDD  Unknown               Unknown  Unknown
testtables         0000000000402A99  Unknown               Unknown  Unknown

See attached buildpack log files for 19.1.2 and 19.1.3.

As a result, specify (and upload) 19.1.2 for ROPP7.0.

[Ian Culverwell]

buildpack mobufr ifort 19.1.2

[Ian Culverwell]

buildpack mobufr ifort 19.1.3

[Dave Offiler]

First look:

According to the (19.1.3) logfile, the BUFR tables are being installed thus:

mkdir: created directory `/data/local/idculv/ROPP/data/bufr/'
`../metdb/TABLEB' -> `/data/local/idculv/ROPP/data/bufr/TABLEB'
`../metdb/TABLED' -> `/data/local/idculv/ROPP/data/bufr/TABLED'
`../metdb/CODEFIG' -> `/data/local/idculv/ROPP/data/bufr/CODEFIG'

but the test programs are looking for:

/data/local/idculv/ROPP/ROPP_tarballs/ropp-7.0/bufr-19.1.3/extra/TABLED

which is PWD when these tests are run. In this release, TABLEB/D are zero-length dummy files (the ones in the metdb/ kernel package are up to date and the ones in extra/ are not to overwrite them), hence the 'end-of-file' error (and not 'file not found').

For some reason, the BUFR_LIBRARY environment variable seems not to be set, has been unset or set to a null value, so the program is not picking up the proper install path. Just why needs more investigation, but a possibility is that I have BUFR_LIBRARY defined in my .profile and Ian doesn't, so I may be using a fall-back path, as I've never seen this problem when testing or installing 19.1.3 for ROPP.

The CHANGELOG for 19.1.3 (Incorrectly labelled 19.1.2) notes:

19.1.2 (3 Jul 2013)
 - Updates to extra/ORIGCENTRE:
   Added Sub-centre code 040 (NOAA/ESRL) to Originating Centre 74 (MetO)
   Added Sub-centre code 041 (WUEL)      to Originating Centre 74 (MetO)
   Added Sub=centre code 000 (NOAA/ESRL) to Originating Centre 59 (NOAA)

 - f90_compilers.dat supports 'ifort13' (Intel F95 XE-2013) on Linux.

 - No changes to any library code or other tables in this release.

but I'd need to diff the buildbufr script and Makefiles to confirm if any 'minor and transparent' changes were made from 19.1.2 but not documented in the CHANGELOG which might explain the error.

[Dave Offiler]

I can't reproduce this on the MetO Linux desktop nor on my home Linux (OpenSUSE 13.1) system with the BUFR stuff completely wiped and re-built clean.

Given that ROPP-7.1 shipped with v19.1.2 and ROPP-8 should work with v20.0.1 (or later), the v19.1.3 release (which is of course not a ROPP component nor PES, but an external 3rd party package) is not going to be used, so this ticket is a "won't fix".