Rename compress parameter

[Dreamsorcerer]

The request() method has a compress parameter which forces compression even when the headers are not set, leading any normal server to not know the request is compressed.

This is a useless option for most people. If we can't remove the option, we should atleast rename it to something that stops people using it by mistake.

We've had many users report bugs due to this confusion.

[wfatherley]

Tracing the parameter starting at ClientSession._request, it's passed to ClientRequest:

aiohttp/aiohttp/client.py

Line 572 in 1d170d3

req = self._request_class(

where it's stored as an attribute. It's then passed to the request object's writer here:

aiohttp/aiohttp/client_reqrep.py

Line 639 in 1d170d3

if self.compress:

Seems that this if-block might be the last point at which the parameter can be interpolated into a Content Encoding header. Because ClientSession and the corresponding objects it interacts with support other parameters that are essentially header values, it doesn't feel unreasonable to keep it around so long as it's not a headache for users. One thing that is worth discussing before implementing a fix is would the value of the parameter compress have precedence over a Content Encoding header if both were passed in?

[Dreamsorcerer]

Hmm, maybe I misread it...
It does set a header at:

aiohttp/aiohttp/client_reqrep.py

Lines 438 to 441 in 1d170d3

	elif self.compress:
	if not isinstance(self.compress, str):
	self.compress = "deflate"
	self.headers[hdrs.CONTENT_ENCODING] = self.compress

Any chance you could create an xfail test based on one of the bug reports?
See for example: #3413

[wfatherley]

So there’s no worry on the edge case of setting the compress attribute after the call to __init__? I was going to suggest that the header-bound parameters reside in descriptor-based attributes, which could circumvent the ceremonial update_* methods…

I’ll have a look at that issue and see what I can come up with :)

[Dreamsorcerer]

So there’s no worry on the edge case of setting the compress attribute after the call to __init__?

Hadn't really thought about that. Not sure why anyone would change it via attribute. But, if that's a problem, we can easily just use @property to call that method when it's changed (or just make it read-only in future).

[Dreamsorcerer]

Might already be a related test in #5478.

[Dreamsorcerer]

Actually, if the test is corrected to use the correct method, then it seems to reproduce the problem when using the C parser. But, works fine with the Python parser...