Print out key giving ValidationError

[GraemeWatt]

The jsonschema package only prints out the value giving a ValidationError, not the associated key. This is often not informative, for example, if the value is None:

error - None is not of type 'string' in None

arising from a submission.yaml file containing:

data_license: {description: null, name: null, url: null}

This line was originally written by default when converting from the oldhepdata format (see HEPData/hepdata-converter#32), therefore many old submissions are now invalidated. The key giving rise to a ValidationError should be printed in addition to the invalid value. See Handling Validation Errors in the jsonschema docs.

[alisonrclarke]

Pasting the above into one of the test files and stepping through, it looks like the path of the ValidationError gives us the details we need:

['data_license', 'name']

So perhaps we could just add the contents of the path, separated by colons or slashes, to the 'in' part of the message, e.g.

error - None is not of type 'string' in data_license:name:None

For an invalid keyword this could look something like:

'energies' is not one of ['cmenergies', 'observables', 'reactions', 'phrases'] in keywords:2:name:energies

(2 is the index of the item in the list - would we want to modify indices to make them easier to read?)

[GraemeWatt]

Yes, that looks good. A simpler approach might be just to use the string representation (source code) of the ValidationError exception, although it's a bit verbose. I tried replacing message=ve.message + ' in ' + str(ve.instance) with message=str(ve) in this line. The error message for null values in the data_license is:

error - None is not of type 'string'

Failed validating 'type' in schema['properties']['data_license']['properties']['name']:
    {'maxLength': 256, 'type': 'string'}

On instance['data_license']['name']:
    None

For an invalid energies keyword the corresponding message is:

error - 'energies' is not one of ['cmenergies', 'observables', 'reactions', 'phrases']

Failed validating 'enum' in schema['properties']['keywords']['items']['properties']['name']:
    {'enum': ['cmenergies', 'observables', 'reactions', 'phrases'],
     'maxLength': 128,
     'type': 'string'}

On instance['keywords'][3]['name']:
    'energies'

[alisonrclarke]

Draft PR in #31. Sample error messages:

Using a number instead of a string for data_file:

error - 12321 is not of type 'string' in 'data_file' (expected: {'type': 'string'})

Missing license info:

error - None is not of type 'string' in 'data_license.name' (expected: {'type': 'string', 'maxLength': 256})

Invalid cmenergies value (this had to be fudged slightly as it doesn't come directly from the schema):

error - Invalid value (in GeV) for cmenergies: '7000 GeV' in 'keywords[2].name.cmenergies' (expected: {'type': 'number or hyphen-separated range of numbers e.g. 1.7-4.7'})

Do these look OK?