Add the active language to generate the cache key

[aaloy@…]

If you use a cache per parge approach and i18n, and you use a cookie to change your default language, if the page are cached previously you have some chances to get the wrong translation.

lang = translation.get_language()
cache_key = 'views.decorators.cache.cache_header.%s.%s_%s' % (key_prefix, iri_to_uri(request.path),lang)

[aaloy <aaloy@…>]

a patch with the proposed cache key

[Jeremy Dunck]

aaloy: I don't think all keys should be affected by i18n. Perhaps there should be an alternate to the low-level cache that does add the active lang as a prefix?

[Jacob]

This should probably only be done when USE_I18N == True, but otherwise this is a good idea.

[Andrew Badr]

This patch works against trunk and checks the USE_I18N setting. However, this patch is current blocked by #730. Incoming requests need to use the Session middleware, then Locale, then Cache, so that the language is determined by the time we hit the cache middleware. However, outgoing responses need Session before Cache as well so that the cookie-based Vary headers can be set.

[mrts]

Can someone please test attachment:middleware_ordering_with_path-weight_tuples-backwards-compatible.diff:ticket:730​ ? Also see comment:26:ticket:730 .

[mrts]

Hrm, Django trac doesn't read links properly...

Patch: ​http://code.djangoproject.com/attachment/ticket/730/middleware_ordering_with_path-weight_tuples-backwards-compatible.diff
Comment: ​http://code.djangoproject.com/ticket/730#comment:26

[Andrew Badr]

Note that this is viable again after [8260]. Patch probably still works, but I didn't like the duplication of code, and it still needs tests and docs of course.

[aaloy]

I'm working on a patch which could work as:

    cache_key = 'views.decorators.cache.cache_header.%s.%s' % (
                      key_prefix, iri_to_uri(request.path))
    if settings.USE_I18N:
       cache_key += ".%s" % translation.get_language()

I have tested using other ways of string concatenations and this one is the fast way (0.6 s for 10^{6 interactions) for the kind of keys we're generating.}

[aaloy]

cache pacth to use lang in the cache key

[aaloy]

patch and tests for cache i18n key generation

[aaloy]

Added the patch an the tests to check that we generate the right keys and that middleware recovers different content depending on the language.

[mrts]

/me wonders how the heck did my totally unrelated comments end up in this ticket!? My guess: nazi transgender aliens did it (it's still April 1 in my timezone, so don't you dare to disagree!).

[Ramiro Morales]

Updated patch to port-r10335 and simplified tests

[Ramiro Morales]

I've updated Antoni Aloy's patch to apply cleanly after r10335 and simplified tests to only include what's needed to run them as part of the Django test suite, also fixed them so they don't break other tests leaving test environment umodified.

[Flo Ledermann]

Any progress on checking this in? This is looking good, isn't it?

[yml]

update the patch against the latest trunk r12453

[Jannis Leidel]

I'm not sure if we break a promise changing the cache key in 1.2 or if it's ok since _generate_cache_header_key is undocumented API anyway.

Anyway, a change needs to be documented.

[aaloy]

jezdez in my oppinion this is not a new feature, is just a bug correction, as one may expect that on a multilingual content page the cache should depend also on the language. Anyway I would be glad to document it.

[Jannis Leidel]

Please don't close the ticket if there wasn't an actual commit.

[aaloy]

Opps. Sorry it was not my intention :(

[aaloy]

Patch documentation

[aaloy]

Added an updated patch with documentation warning about the new cache key generation. As is an internal feature I'm not really sure about if it has to be documented or not.

[Jannis Leidel]

(In [12546]) Fixed #5691 - Adds the active language to the cache key. Thanks, Antoni Aloy, Ramiro Morales and Yann Malet.

[Jacob]

Milestone 1.2 deleted

[Aymeric Augustin]

In [17061]:

Made the cache locale-dependant when USE_L10N is True, even if USE_I18N is False. Refs #5691.

[Mariusz Felisiak <felisiak.mariusz@…>]

In 258c88a:

Refs #5691 -- Made cache keys independent of USE_L10N.

This mostly reverts af1893c4ff8fdbf227a43a559d90bb1c1238b01a.