Arch Linux User Repository

Hm, the source was unchanged upstream. Only the test tarball changed so I needed to upload an updated package.

That said, maybe Manjaro uses a different Compiler or Compiler-Version? I don't get thia issue on a Clean Arch- Container.

Best, Jan

Hello, Does anybody understand this error message while attempting to build the latest version (2.48.0-2) on a Manjaro system:

data/software/AUR/eccodes/src/eccodes-2.48.0-Source/src/eccodes/grib_dumper_factory.cc:19:8: warning: type ‘struct table_entry’ violates the C++ One Definition Rule [-Wodr]
   19 | struct table_entry
      |        ^
/data/software/AUR/eccodes/src/eccodes-2.48.0-Source/src/eccodes/grib_iterator_factory.cc:18:8: note: a different type is defined in another translation unit
   18 | struct table_entry
      |        ^
/data/software/AUR/eccodes/src/eccodes-2.48.0-Source/src/eccodes/grib_dumper_factory.cc:22:23: note: the first difference of corresponding definitions is field ‘dumper’
   22 |     eccodes::Dumper** dumper;
      |                       ^
/data/software/AUR/eccodes/src/eccodes-2.48.0-Source/src/eccodes/grib_iterator_factory.cc:21:39: note: a field with different name is defined in another translation unit
   21 |     eccodes::geo_iterator::Iterator** iterator;
      |                                       ^
[ 44%] Built target eccodes
make: *** [Makefile:166: all] Error 2
==> ERROR: A failure occurred in build().
    Aborting...

I get random failures on my test pipeline, too. though not too many and not too often. It might be a good idea to report such failures upstream so the maintainers can check and possibly fix things, especially when tests fail permanently on a specific machine.

It also might be a good idea to check the error persists in a clean chroot (see: https://wiki.archlinux.org/title/DeveloperWiki:Building_in_a_clean_chroot for details).

Best Jan

Can confirm eccodes_t_grib_mtg2_switch failing for quite a while now, always ending up in the output below. Over here, this is the case for 2.41.0-1 through 2.47.0-1 (2.41.0-1 being the last working version).

edit: FWIW, if you look at the source and compare tags 2.41.1 and 2.42.0, there were a lot of changes in grib_mtg2_test/grib_mtg2_switch_test, CASE 23: CERISE ECMWF pre-mtg2 being introduced among others. Ever since, this has been failing over here.

Hi, one of the tests failed for me when trying to install version 2.46.0-3:

The following tests FAILED:
     24 - eccodes_t_grib_mtg2_switch (Failed)               eccodes sanity script

Here are the last lines of the output for that test:

+ for i in "${!cases[@]}"
+ grib_check_key_equals temp.grib_mtg2_switch_test.case23.grib2 MTG2SwitchViaTablesVersion,MTG2SwitchDefault,MTG2Switch '0 1 0'
+ a_file=temp.grib_mtg2_switch_test.case23.grib2
+ a_key=MTG2SwitchViaTablesVersion,MTG2SwitchDefault,MTG2Switch
+ a_expected='0 1 0'
++ /home/valentina/.cache/yay/eccodes/src/build/bin/grib_get -p MTG2SwitchViaTablesVersion,MTG2SwitchDefault,MTG2Switch temp.grib_mtg2_switch_test.case23.grib2
+ a_result='1 0 0'
+ '[' '1 0 0' '!=' '0 1 0' ']'
+ echo 'File:     '\''temp.grib_mtg2_switch_test.case23.grib2'\'''
File:     'temp.grib_mtg2_switch_test.case23.grib2'
+ echo 'Key(s):   '\''MTG2SwitchViaTablesVersion,MTG2SwitchDefault,MTG2Switch'\'''
Key(s):   'MTG2SwitchViaTablesVersion,MTG2SwitchDefault,MTG2Switch'
+ echo 'Expected: '\''0 1 0'\'''
Expected: '0 1 0'
+ echo 'Result:   '\''1 0 0'\'''
Result:   '1 0 0'
+ exit 1

Corresponding case in the test script seems to be this one: https://github.com/ecmwf/eccodes/blob/c10e8be0c3e27dd2be36a909b06d5d3be6889708/tests/grib_mtg2_switch.sh#L619

It does not seem to me that this is caused by something local to my machine but I might be wrong.

Best, Valentina

Hi Jan, Thanks for this update. I confirm it works for me too now.

During compilation, I observed some ODR violations warnings (see below), perhaps this explains that. Maybe this should be reported upstream ?

/home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source/src/eccodes/grib_hash_keys.cc:11739:8: warning: type ‘struct grib_itrie’ violates the C++ One Definition Rule [-Wodr]
11739 | struct grib_itrie
      |        ^
/home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source/src/eccodes/grib_itrie.cc:278:8: note: a different type is defined in another translation unit
  278 | struct grib_itrie
      |        ^
/home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source/src/eccodes/grib_hash_keys.cc:11741:17: note: the first difference of corresponding definitions is field ‘next’
11741 |     grib_itrie* next[SIZE];
      |                 ^
/home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source/src/eccodes/grib_itrie.cc:280:17: note: a field of same name but different type is defined in another translation unit
  280 |     grib_itrie* next[SIZE];
      |                 ^
/home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source/src/eccodes/grib_hash_keys.cc:11739:8: note: array types have different bounds
...

Hi @Nestor_013,

I get this very error on my local machine, too, but not on my test pipelines using fresh Arch-containers. I'm not really sure what cases the problem, but I deactivated the test and re-added the download of the extra test data since it really speeds up the test process. An update of the package is on the way.

Best Jan

Hi, not sure if it's me only, but I get a crash, which prevents yay from installing, in one of the tests (see below):

 91/462 Test  #91: eccodes_t_grib_ecc-2221 ........................***Failed    0.12 sec
++ echo 'Script: /home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source/tests/grib_ecc-2221.sh'
Script: /home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source/tests/grib_ecc-2221.sh
++ proj_dir=/home/user/.cache/yay/eccodes/src/eccodes-2.46.0-Source
++ build_dir=/home/user/.cache/yay/eccodes/src/build
++ data_dir=/home/user/.cache/yay/eccodes/src/build/data
++ def_dir=/home/user/.cache/yay/eccodes/src/build/share/eccodes/definitions
++ ECCODES_DEFINITION_PATH=/home/user/.cache/yay/eccodes/src/build/share/eccodes/definitions
++ export ECCODES_DEFINITION_PATH
++ tools_dir=/home/user/.cache/yay/eccodes/src/build/bin
++ bin_dir=/home/user/.cache/yay/eccodes/src/build/bin
++ EXEC=
++ test x '!=' x
++ test_dir=/home/user/.cache/yay/eccodes/src/build/tests
++ samp_dir=/home/user/.cache/yay/eccodes/src/build/share/eccodes/samples
++ ECCODES_SAMPLES_PATH=/home/user/.cache/yay/eccodes/src/build/share/eccodes/samples
++ export ECCODES_SAMPLES_PATH
++ set -u
++ HAVE_PRODUCT_BUFR=1
++ HAVE_PRODUCT_GRIB=1
++ HAVE_JPEG=1
++ HAVE_LIBJASPER=0
++ HAVE_LIBOPENJPEG=1
++ HAVE_PNG=1
++ HAVE_AEC=1
++ HAVE_GEOGRAPHY=1
++ HAVE_ECKIT_GEO=0
++ HAVE_EXTRA_TESTS=1
++ HAVE_MEMFS=0
++ ECCODES_ON_WINDOWS=0
+++ pwd
++ echo 'Current directory: /home/user/.cache/yay/eccodes/src/build/tests'
Current directory: /home/user/.cache/yay/eccodes/src/build/tests
+ label=grib_ecc-2221_test
+ tempGrib=temp.grib_ecc-2221_test.grib
+ tempFilt=temp.grib_ecc-2221_test.filt
+ sample_grib2=/home/user/.cache/yay/eccodes/src/build/share/eccodes/samples/GRIB2.tmpl
+ cat
+ /home/user/.cache/yay/eccodes/src/build/bin/grib_filter -o temp.grib_ecc-2221_test.grib temp.grib_ecc-2221_test.filt /home/user/.cache/yay/eccodes/src/build/share/eccodes/samples/GRIB2.tmpl
+ grib_check_key_equals temp.grib_ecc-2221_test.grib model,configuration,forcing,timespan 'lisflood v5 ecmf-ifs 6h'
+ a_file=temp.grib_ecc-2221_test.grib
+ a_key=model,configuration,forcing,timespan
+ a_expected='lisflood v5 ecmf-ifs 6h'
++ /home/user/.cache/yay/eccodes/src/build/bin/grib_get -p model,configuration,forcing,timespan temp.grib_ecc-2221_test.grib
ECCODES ERROR   :  timespan (Key/value not found)
+ a_result='lisflood v5 ecmf-ifs '

Hi Jan, looks like the certificate on the Arch Wiki page has expired so I'll have to defer attempting the clean Chroot build.

I am, however, by no means certain that (for all the portentousness of the "fatal error") that the Fortran fail is the final cause of failure as I don't see a connection between that and the C++ errors of the second block, and the build continues far longer than any other "waiting for unfinished jobs" that I've ever seen.