[Question] is it possible to run separate hooks for `dumps` and `unstructure`

[amirreza-sf80]

hi 👋
sorry i ask too much🙏

so the issue is with this code:

import uuid

from bson.binary import Binary

from cattrs.preconf.bson import make_converter


converter = make_converter()

converter.register_unstructure_hook(uuid.UUID, lambda x: Binary.from_uuid(x))
converter.register_unstructure_hook(
    Union[uuid.UUID, None], lambda x: Binary.from_uuid(x) if x else None
)

bson doesn't serialize UUIDs
but they have Binary.from_uuid(x) so we can achive it like this

and it works

s = bson_converter.structure({"uu": uuid.uuid4()}, Something)
dumps = bson_converter.dumps(s)

but when doing bson_converter.unstructure(s), i would expect to see a UUID object
but since dumps and unstructure use the same hooks, i see this instead
'uu': Binary(b'\xd5\xd2\x0e\x15TWJ\xf0\xbae\xe6\xed\x8a\xc6\xa9\xb3', 4)

which is not a big issue for me, but it got me thinking if there is a workaround

[Tinche]

.dumps() is just a thin wrapper around .unstructure() and whatever the library's version of dumps() is, so not really.

Generally when using a preconf converter, I would expect to almost always use dumps(). Their unstructure() output can be significantly different than what you would get with an ordinary converter.

For example to illustrate the point, the msgspec converter will leave certain dataclasses completely untouched when unstructuring, because msgspec can handle them faster. But we only do this if the output makes sense for cattrs semantics; for example msgspec skips private fields and we do not by default, so if a class has a private field, we handle it instead.

If you need two different paths, I would use two different converters (they are relatively cheap). You can even .copy() a converter so you don't have to apply common customizations twice.

[amirreza-sf80]

thank you 🙏