On Thu, October 26, 2006 2:44 pm, Joe Orton wrote: > This is a non-starter. The cache cannot rely on any particular > implementation detail of the trunk core output filter (which sounds like > a bug), nor can it assume that filters between it and the core output > filter will pass through FILE buckets intact - think about e.g. SSL.
In that case the solution is to expose the logic inside ap_core_output_filter() that decides whether the socket will block, and make that generally available. I have a better solution for the split-large-brigades concept inspired by your patch from yesterday - there is a placeholder for this in there. Regards, Graham --
