modify FDB env var checks

[cosunae]

This PR tries to solve two issues:

• FDB library can be set through two env variables, FDB_HOME or FDB5_HOME. earthkit-data requires here FDB_HOME but pyfdb requires FDB5_HOME: https://github.com/ecmwf/pyfdb/blob/b004bd58ed18c3d7e8bfc3e8081ae6c24fc4b739/pyfdb/pyfdb.py#L44
  it would be better if all python ecosystem using FDB would require the same variable, therefore I propose here to use FDB5_HOME to be compatible with pyfdb.
• The config can be set through FDB5_CONFIG or FDB5_CONFIG_FILE.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.10%. Comparing base (045d5d7) to head (162deb9).

Additional details and impacted files

@@           Coverage Diff            @@
##           develop     #414   +/-   ##
========================================
  Coverage    90.10%   90.10%           
========================================
  Files          121      121           
  Lines         7905     7905           
  Branches       679      679           
========================================
  Hits          7123     7123           
  Misses         651      651           
  Partials       131      131           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[cfkanesan]

FDB5_DIR can also be used to locate the library

[sandorkertesz]

Hi @cosunae, thank you for this PR.

I agree with you that the checking for FDB support should be more consistent. However, please consider these points:

1. pyfdb finds the fdb5 library by using findlibs:

    libName = findlibs.find("fdb5")

   This call tries to find it by applying the algorithm described at: https://github.com/ecmwf/findlibs/tree/develop. It checks many things e.g. env variables like FDB5_HOME, FDB5_DIR etc, but it also checks e.g. the conda env.

2. The checks in sources/fdb.py are meant to test the availability of the FDB config rather than that of the library, since if the library is not available loading sources/fdb.py will raise an exception immediately (see atop of the file).

3. The checks you propose are too strict because with this I need both FDB5_HOME and FDB5_CONFIG/FDB5_CONFIG_FILE to be defined.

    if fdb5_home is None or (fdb5_conf is None and fdb5_config_file is None):
               raise ValueError(

   However, at the moment I am able to use pyfdb in my conda env (containing the FDB5 lib) having only FDB_HOME (!) defined.

[cosunae]

Thanks @sandorkertesz
let me reply to your comments:

1. yes, my understanding of the find_libs algorithm is that it will mainly check for FDB5_XXX instead of FDB_XXX, therefore the reason for this proposal. Is this assumption right ?
2. understood, thanks for the clarification.
3. I see, yes, if FDB5_HOME is enough we should not make it stricter. I changed it to have any of the variables defined, can you check now ?

[sandorkertesz]

3. I see, yes, if FDB5_HOME is enough we should not make it stricter. I changed it to have any of the variables defined, can you check now ?

The new check seems to be fine now! However, internally we do use FDB_HOME, so I suggest you add it back and check its existence together with the other 3. Then this PR can be merged.