Add more careful norm_text check

[cthoyt]

This makes sure it's not none before checking if it can be stripped. Response to this error:

...
  File "/Users/cthoyt/dev/gilda/gilda/grounder.py", line 685, in load_entries_from_terms_file
    yield Term(*row_nones)
          ^^^^^^^^^^^^^^^^
  File "/Users/cthoyt/dev/gilda/gilda/term.py", line 55, in __init__
    if not norm_text.strip():
           ^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'strip'

[bgyori]

This probably doesn't hurt, but assuming that norm_text is produced by the normalize function, under what conditions could it be None?

[cthoyt]

It's happening in a place where I'm using load_entries_from_terms_file() - something in that process is leaving a None in the norm_text column. I'm trying to track that down now too

[cthoyt]

Tracked down the culprit by hacking some logging in to the load_entries_from_terms_file() function:

WARNING: [2024-02-22 01:40:56] gilda.grounder - [None, '-', 'umls', 'C5187432', '-', 'synonym', 'umls', None, None, None] Normalized text for Term cannot be None nor empty

https://uts.nlm.nih.gov/uts/umls/concept/C5187432

nice name -

[bgyori]

The odd thing is that normalize('-') returns '', as expected, so the None is likely coming from some other code. Is it clear where that happens?

[cthoyt]

I am thinking this might be a weird syncing issue - I don't have UMLS update on the normal cycle since it's so big and hard to work with. I think what happened is that the UMLS index was dumped before we put the check inside the Term object, and so it's indexed with an empty string at the moment, which the load function turns into none since it does a truthy check. I'll try rebuilding UMLS and see what happens