On 25 Aug 2010, at 07:41, Paul Querna wrote: > Grump old man mode, this seems moderately disturbing. > > Previously it was single hash table look up on the content-type, which > contained a list of all filters to apply.
With lots of brokenness, discussed quite a few times over the years. This was a long-overdue round tuit. PR 33499 isn't the best reference, but is the first one to come up on a bugzilla search. > It has been replaced with iterating all possible filters and executing > an the expression parser for each possible filter. You make that sound a lot. How many lookups can you envisage in any case that *would work at all under the old implementation*? > Is anyone else surprised at this change? I did a quick search on > dev@, and I couldn't find any previous discussion of this change.... IIRC it was first suggested as an alternative to removing AddOutputFilterByType altogether from 2.1/2.2, over five years ago. The underlying problem: checking content-type at the wrong time (before we knew it for dynamic or proxied contents), and the regular likelihood of multiple calls to ap_set_content_type per request. -- Nick Kew