Improve custom error handling.

[chadrik]

The Schema 'error' kwarg now supports a callable, which returns the error message as a string. Its arguments are:

• schema: the underlying schema object
• data: the data being validated
• reason: a string constant for the type of error encountered
• details: further details about the reason for the failure. may be an exeption, a string, or None

Additionally, error strings passed to the Schema 'error' kwarg are now provided more context about the error in the form of four {fields} for use with "new-style" string formatting. These are string representations of the four arguments described above.

[chadrik]

One design decision that I've been wavering on is whether the error callable should be passed 4 arguments, like so:

def auto_error(schema, data, reason, details):
    return 'Error encountered with {data!r}'.format(data=data)

or if it should take a dictionary so that it will be easier to add more context items later without breaking existing error handlers:

def auto_error(context):
    return 'Error encountered with {data!r}'.format(**context)

[chadrik]

I just noticed the doctest is failing on Travis.

The reason for the error is that inside the new format_error function I am formatting all callable schemas using their __name__ attribute, if it exists:

    if callable(schema) and hasattr(schema, '__name__'):
        schema = schema.__name__
    else:
        schema = repr(schema)

This means that errors that used to look like this:

SchemaError: '123' should be instance of <type 'int'>

Now look like this:

SchemaError: '123' should be instance of int

Personally, I prefer the latter, but if you don't want to make this change, I can modify the stringification of callables to only use __name__ for callable functions, like so:

    if inspect.isfunction(schema) and hasattr(schema, '__name__'):
        schema = schema.__name__
    else:
        schema = repr(schema)

Let me know what you prefer.

[erikrose]

I feel like these should be different subclasses of SchemaError. Then we could catch them selectively with try/except in calling code. What do you think?

[chadrik]

I think that I made this pull request over a year ago and it's obvious that this project is not actively maintained. I've moved on. schemtatics looks interesting....