Allows Content-Length for 304 Not Modified response

[alefranz]

Resolves #2099

[halter73]

You're getting good at this 😄

[halter73]

@Tratcher Care to review? Feel free to merge.

[Tratcher]

Nice and simple

[halter73]

@Tratcher I was thinking is that maybe we should verify that the actual length of the response body is zero for 304 responses instead of just skipping the response-body-too-short validation.

I didn't bring this up in the PR review because it's a separate issue and we'll probably want to do the same validation for HEAD requests. Do you think this is worth addressing?

[halter73]

I'm not too familiar with the spec requires wrt this. Is a 304 response ever allowed to have a non-empty response body?

[Tratcher]

304 is pretty special, it's a lot like a head request. It explicitly must not have a response body, but it is allowed to describe the body it would have had if it were a 200.

[alefranz]

@halter73 I thought about the response length validation as well, but as you said is a separate issue. Also it can be considered a breaking change.
I'm happy to have a look if I find a clean way to implement it.
Should I open a issue for discussion?

[halter73]

We're good then. We already log and ignore response body writes for HEAD requests and 204/205/304 responses.

I'm assuming we should not allow a non-zero content length to be set for 204 and 205 responses.

[halter73]

See here for the logic that decides whether or not to ignore writes to the response body.

[alefranz]

304 is include, see here
and it's already handled as write throws if it can't have body.
Tests here

Nothing else to be done, right?

[halter73]

I think we have everything covered. I just wanted someone to agree that allowing a non-zero Content-Length response header for 204 and 205 responses is silly.