User messages for missing files

[toastisme]

This adds a warning for dials.import if no input files are found when using a wildcard, and an error message if a specific requested file is missing.

Currently the behaviour when using wildcards is to show help if no files are found, or when using multiple wildcards to show nothing if some wildcards return nothing. This adds a warning to the user if a wildcard returns nothing.

The current behaviour for when a specific file is not found is to show the traceback from dxtbx. This pr hides the traceback and shows just the file not found error, followed by calling help.

[codecov]

Codecov Report

Merging #1863 (34ff460) into main (120cd59) will decrease coverage by 0.00%.
The diff coverage is 66.66%.

@@            Coverage Diff             @@
##             main    #1863      +/-   ##
==========================================
- Coverage   80.36%   80.36%   -0.01%     
==========================================
  Files         580      580              
  Lines       65698    65706       +8     
  Branches     9259     9260       +1     
==========================================
+ Hits        52800    52805       +5     
- Misses      10844    10847       +3     
  Partials     2054     2054              

[rjgildea]

Might be useful to add a simple test case to cover this (although possibly overkill)?

[toastisme]

Might be useful to add a simple test case to cover this (although possibly overkill)?

Not quite sure what this would be testing if it's just logger statements based on Glob and a FileNotFoundError. I guess you could have a test to confirm that trying to import a non-existent file always gives a logger.error followed by calling help, and trying to import multiple wildcards where one returns nothing gives only a logger.warning. Does feel a bit overkill to me..

[ndevenish]

I think part of the confusion is that control flow here is rather non-obvious - this prints the message, but doesn't mark it as unhandled. In dials.import this is only called to parse images, so no experiment is ever created. dials.import then sees that there are no experiments, and prints the help.

I don't think printing help is particularly helpful in this case, because there was (otherwise looking valid) input, and the giant block of text somewhat hides the actual error message. If the user asked for a file and it wasn't present, this seems a somewhat reasonable case to call sys.exit()?

What other downstream scenarios are there that a missing file here is a recoverable error?

(Otherwise, this is a measurable improvement over a stack trace)

[toastisme]

Sorry this had slipped my mind. Yes I agree the solution isn't ideal. It would be more consistent for missing files to be caught in unhandled. My original reason for not doing this was that it's not particularly helpful for the user to see an argument is unhandled but not know it's due to missing files.

A better solution might be changing unhandled arguments in general to be a dictionary which explain the reason for each unhandled argument (99% of which I'm guessing would just be unrecognised arguments).

[ndevenish]

Even though I think we could do better here, this is absolutely an improvement to what we had before.