Raise TypeError on invalid query params.

[lovelydinosaur]

Prompted by #2515. (Doesn't presume any particular resolution to that case.)

Closes #746 - we don't support bytes (or other unexpected types) as QueryParam values. Let's raise a clear error, rather than coercing an unexpected str value.

Before this change...

>>> httpx.URL("http://www.example.com/", params={"query": b"abc"})
URL('http://www.example.com/?query=b%27abc%27')

After this change...

>>> httpx.URL("http://www.example.com/", params={"query": b"abc"})
TypeError: Expected str, int, float, bool, or None. Got <class 'bytes'>.

[zanieb]

Seems nice to ensure that the formatted message is correct

Suggested change

	with pytest.raises(TypeError):
	with pytest.raises(TypeError, match=r"Expected str, int, float, bool, or None\. Got 'bytes'\."):

[lovelydinosaur]

That's neat, thanks.

[zanieb]

Sorry I broke your line length :)

[lovelydinosaur]

🤨

[zanieb]

You don't need the r if you aren't escaping the periods — this is doing a re.search so technically the . will be a wildcard but 🤷‍♀️ it's not likely to cause you any problems. You could use re.escape if you really wanted to be correct.

[zanieb]

@tomchristie this broke Prefect since we were passing UUID objects — that's fine but I think we should make sure this is listed on the changelog for 0.23.2 for visibility.

[lovelydinosaur]

Ooofff. Okay, more care needed in the future.

[lovelydinosaur]

That's entirely on me - rushing things that didn't need to be rushed.

Do we want to roll this back in a 0.23.3, and leave it for the upcoming 0.24.0 update? (Or do we leave it as is?)

[zanieb]

It'd probably be safest to make this change in 0.24.0 instead since it changes the supported interface (I'll try to keep an eye out for this in future reviews), but I don't have strong feelings about rolling it back.

[curtiscook]

@tomchristie this broke Prefect since we were passing UUID objects — that's fine but I think we should make sure this is listed on the changelog for 0.23.2 for visibility.

FYI - This also breaks datetime -> str conversion -- would suggest to keep datetime supported and adhering to RFC