Handle missing or invalid path input for commands

[juhoinkinen]

A full traceback is shown for trying to e.g. train on a nonexistent path:

$ annif train tfidf-fi nonexistent_path
...
  File "/home/local/jmminkin/git/Annif/annif/corpus/document.py", line 65, in documents
    with opener(self.path) as tsvfile:
FileNotFoundError: [Errno 2] No such file or directory: 'nonexistent_path'

Similarly for trying to train without a path:

$ annif train tfidf-fi
...
  File "/home/local/jmminkin/git/Annif/annif/cli.py", line 53, in open_documents
    docs = open_doc_path(paths[0])
IndexError: tuple index out of range

The first case could be resolved by adding exists=True argument to type=click.Path(). This (at least alone) does not work in the second case, for which maybe required=True could be used, but the Note on Non-Empty Variadic Arguments should be considered.

Similar tracebacks arise when using:

• train
• learn
• eval
• optimize

For some reason index behaves differently:

• It gives proper error when not given a path:

  $ annif index tfidf-fi
  Usage: annif index [OPTIONS] PROJECT_ID DIRECTORY
  Try "annif index --help" for help.
  
  Error: Missing argument "DIRECTORY".
  

• It does not produce any output or error with nonexistent path:

  $ annif index tfidf-fi nonexistent
  $
  

[osma]

+1 for adding exists=True.

The note on required file arguments makes a good point.

I've sometimes (when testing VW based backends that support online learning) used a command like

annif train vw-multi /dev/null

The idea is to create an empty model which can then be trained incrementally.

I think annif train vw-multi (without a file argument) should do the same as the above, but perhaps emit an info message about creating an empty model. I'm unsure what this would do with non-VW backend though - probably there should be an error because creating an empty tfidf or fasttext model doesn't make sense.