Don't consider Accept-Language when generating a cache key

[choongmin@…]

Currently, Django makes different cache keys for different Accept-Language headers. But different Accept-Language headers may be resolved into the same language. For example, see the following Accept-Language headers collected in my site for last 30 minutes:

• en-us
• en-US
• en-US,en;q=0.8
• en-US,en;q=0.8,ko;q=0.6
• fr-fr
• ja-JP
• ko
• ko, en
• ko-kr
• ko-KR
• ko-KR, en-US
• ko-kr,ko;q=0.8,en-us;q=0.5,en;q=0.3
• ko-KR,ko;q=0.8,en-US;q=0.6,en;q=0.4
• ko;q=1.0,en;q=0.5

It makes the cache system inefficient for multilingual sites. Since there is utils.cache._i18n_cache_key_suffix, it is unnecessary and redundant to include Accept-Language headers in a cache key. We should ignore them in utils.cache.learn_cache_key.

Note that not adding "Vary: Accept-Language" from the first place in middleware.locale.process_response is not a good solution because of proxy servers.

The attached patch removes Accept-Language from headerlist before generating a cache key.

If the patch is rejected, users like me should at least be able to ignore them.

[Łukasz Langa]

Pull request:

• ​https://github.com/django/django/pull/851

Remarks:

• tested on both 2.7 and 3.3
• ​http://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.44 doesn't give any meaning to the order of headers listed in Vary so I sorted those headers while creating the cache key to make sure the number of combinations is the smallest possible

[Carl Meyer <carl@…>]

In 6a057e1595dfe45a0411c116f603801266c87395:

Fixed #18191 -- Don't consider Accept-Language redundantly in cache key.

Thanks to choongmin for the original patch.