Gzip Middleware Should Compress 201 (Created) Responses

[rwagner@…]

Hi,

RESTful web services that use POST for creation of resources often respond with a representation of the newly created resource, which can be lengthy. However, the appropriate status code is 201, which the Gzip middleware ignores. There's a trivial patch attached that contains these lines:

-        if response.status_code != 200 or len(response.content) < 200:
+        if response.status_code not in (200, 201) or len(response.content) < 200:

This is not a particularly big deal, just a convenience.

Thanks,
Rick

[Julian Bez]

While we are at it: What about 404 or 403 responses?

[Rick Wagner]

Replying to julianb:

While we are at it: What about 404 or 403 responses?

I don't believe that was what was intended by the original condition. The 2XX status codes indicate successful operations, whereas 4XX are client errors. In practice, I think it is a good idea to render errors in the simplest formats possible, to make interpreting them easier on the client side. But, I can see the argument for compressing a custom 404 response that is rendered from a template, and could get large.

--Rick

[Aymeric Augustin]

The comment in gzip.py say: "It's not worth compressing non-OK or really short responses."

If non-OK responses are not worth compressing because they are expected to be short, testing their length is enough, which means the check on response.status_code should be removed.

If there is another reason why non-OK responses should not be compressed, the proper check is 200 <= response.status_code < 300, and the reason should be explained in a comment.

[Aymeric Augustin]

Regarding Rick's comment above: the content will only be gzipped if the user agent sent an Accept-Encoding: gzip header. It seems reasonable to assume it will be able to decode gzip, even if the server returns an error page.

[Georges Dubus]

The previous patch, with added tests

[Georges Dubus]

Here are simple tests to check that long responses are compressed and short responses are not, regardless of the return type. Are they ok ?

[Aymeric Augustin]

See also #17514.

[Aymeric Augustin]

In [17365]:

Fixed #10762, #17514 -- Prevented the GZip middleware from returning a response longer than the original content, allowed compression of non-200 responses, and added tests (there were none). Thanks cannona for the initial patch.