Throwing an exception out of a custom field (or any QuerySet._results_iter callable) results in a bad _results_cache

[Jeremy Dunck]

Using a custom field which breaks:

class BrokenField(DateTimeField):
    __metaclass__ = SubfieldBase

    def to_python(self, value):
        raise TypeError("Nope.")

Assume some rows exist.

If you then create AModel using that field:

class AModel(Model):
   f = BrokenField(null=True)

And you then try to query on it:

>>> qs = AModel.objects.all()
>>> qs
[]

>>> qs._result_cache is None
True

You get an empty (or truncated) results list.

I think the underlying exception should instead be raised. If it only breaks sometimes, the results will be shorter, a very hard-to-find bug.

[Luke Plant]

There may be some reason why this is hard to do, but I have accepted the bug in principle.

[anonymous]

This seems to be fixed in 1.5 since the _result_cache setup changed