[EMAIL PROTECTED] wrote: > Marc M. Adkins wrote: > >> The ap_hook_handler() call >> does not specify the handler key from the corresponding AddHandler >> configuration directive. As a consequence, the specified handler >> function >> must look at and accept or decline each request. > > > yes, you are right. IMO that sucks, for a number of reasons: > > * the Apache 1.3 core was better about it and only called handlers when > the request type matched a type the handler was interested in, > > * we had to change the module API to accomodate this regression in the > core, and > > * now we have a distributed performance hit in 2.x, because the core is > calling all these modules who don't care about the request which makes > the CPU fetch crap instructions into the i-cache. This won't show up as > a hot spot in a profiler, because it is spread among all the different > handlers and other code which is now taking more frequent i-cache misses. > > ...but I'll shut up because I don't have a patch or a design for one yet. > >> By contrast, output filters only see requests in which they have >> interest. > > > Clearly a better situation.
The change in handlers was made for a reason. The problem is, I've forgotten the reason! A survey of all the existing handlers might well reveal it - otoh, the reason might simply be that filtering hooks didn't fit neatly with the new hook API. Which is, of course, fixable, at the cost of some complication. Cheers, Ben. -- http://www.apache-ssl.org/ben.html http://www.thebunker.net/ "There is no limit to what a man can do or how far he can go if he doesn't mind who gets the credit." - Robert Woodruff
