httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Niklas Edmundsson <>
Subject Re: [PATCH] mod_cache 304 on HEAD (bug 41230)
Date Tue, 17 Apr 2007 07:07:09 GMT
On Mon, 16 Apr 2007, Ruediger Pluem wrote:

>> I have now tested this patch, and it seems to solve the problem. This is
>> on httpd-2.2.4 + patch for PR41475 + our mod_disk_cache patches.
>> Without the patch a HEAD on a cached expired object that isn't modified
>> will unconditionally return 304 and furthermore cause the cached object
>> to be deleted. We believe that this is the explanation to why it has
>> been so hard to track down this bug - it only bites one user and that
>> user usually has no clue on what happens, and even if we try to
>> reproduce it immediately afterwards it won't trigger.
>> With the patch stuff works like expected:
>> - A HEAD on a cached expired object that isn't modified will update
>>   the cache header and return the proper return code, it follows the
>>   same code path as other requests on expired unmodified objects.
>> - A HEAD on a cached expired object that IS modified will remove the
>>   object from cache and then decline the opportunity to cache the
>>   object.
> Are you really sure that it gets deleted? cache->provider->remove_entity does
> not really remove the object from the cache. Only cache->provider->remove_url
> does this.

Yes, but the CACHE_REMOVE_URL filter will remove it, right? It removes 
the CACHE_REMOVE_URL filter only after it has decided that it's 
actually caching the response so it will bite in that case.

> I consider the CACHE_SAVE filter already as hard to read (not your 
> fault by any means), but from my point of view your patch does 
> increase this (specificly I think about the rv = !OK line. I know 
> that a similar trick is done some lines above, but I don't like that 
> one either).

I also found rv=!OK ugly, but I just followed the established style to 
create a minimal patch without extra fuzz. Feel free to clean stuff up 
to improve readability, as long as the bug gets fixed I'm happy :)

  Niklas Edmundsson, Admin @ {acc,hpc2n}      |
  Push any key. Then push the "any other" key.

View raw message