snapshottest: 0.5.1 Breaks tests

@StoicLoofah thanks for tracking this down.

I think there is a snapshottest bug or mis-feature in CollectionFormatter.normalize:

class CollectionFormatter(TypeFormatter):
    def normalize(self, value, formatter):
        iterator = iter(value.items()) if isinstance(value, dict) else iter(value)
        # ... This line is the problem:
        return value.__class__([formatter.normalize(item) for item in iterator])

This is going to have problems with all collection subclasses that require additional parameters in their constructors. (And indeed, there’s already an override for defaultdict.)

At a minimum, CollectionFormatter.normalize should probably catch exceptions trying to re-construct a normalized object, and issue a more helpful error message (suggesting the user either convert the snapshotted object to a simpler type, or that they add a custom formatter). Ideally, snapshottest would also preflight this while writing snapshot files, so that the error is reported then, not only when comparing snapshots later.

But I’m wondering if a better fix might be to change the value.__class__(...) line to instead construct the base type the formatter writes into snapshots. E.g., CollectionFormatter(dict, format_dict).normalize(value) should just try to construct a plain dict—even if value is a DRF ReturnDict—because format_dict always just writes snapshots as plain dicts. Then we wouldn’t need all these extra custom formatters: if a formatter is willing to write a snapshot, then it really should be able to compare the same type of object to that snapshot later.

(I confess I’m getting a little lost in understanding the need for this normalize logic in the first place. It transforms the actual value being compared (not the expected snapshot value) prior to comparison. I think this was added in #82 to fix problems introduced in a bunch of other major changes between v0.5.0 and v0.5.1.)

Also, it seems like the original report from @hobochild is probably related, but I think that specific case might be covered by the fix in #116 that went into v0.6.0.

I was able to fix it by adding the following into my conftest.py

from rest_framework.utils.serializer_helpers import ReturnDict, ReturnList
from snapshottest import formatter, formatters

class DrfReturnCollectionFormatter(formatters.TypeFormatter):
    def normalize(self, value, formatter):
        iterator = iter(value.items()) if isinstance(value, dict) else iter(value)
        # Added the serializer kwarg as required by ReturnDict or ReturnList
        return value.__class__([formatter.normalize(item) for item in iterator],
                               serializer=None)

formatter.Formatter.register_formatter(DrfReturnCollectionFormatter(ReturnDict,
                                                                    formatters.format_dict))
formatter.Formatter.register_formatter(DrfReturnCollectionFormatter(ReturnList,
                                                                    formatters.format_list))

Not sure how you guys want to handle this since it’s an upstream issue but figured I would let you know how I addressed it!