On Tue, Jun 08, 2010 at 05:20:16PM +0200, "Plüm, Rüdiger, VF-Group" wrote: > Is it possible to have any non subrequest specific filters below the > ap_subreq_core_filter_handle? > It is no that the current patch does not handle this but wouldn't it be > possible > to just throw away any filter between > > r->output_filters and ap_subreq_core_filter_handle without any further test?
I'd spent some time working on exactly that approach, trying to answer that question. But surprisingly the answer is "yes" - the subreq filter has ftype=AP_FTYPE_CONTENT_SET, so any filter registered with an ftype greater than that can be inserted and will work (so long as it doesn't depend on seeing EOS). So I couldn't convince myself it was correct to leave such filters in place, though the situation is certainly odd. A solution to this problem which André talked about in the list archive was to turn the subreq core filter into a "chaining" filter, perhaps at PROTOCOL level within the subreq, to ensure it is "really last" in the chain. This seems cleaner than the current mess, but I couldn't get it to work without regressions (particularly in fast internal redirect handling). Regards, Joe
