On Sat, Sep 13, 2014 at 9:23 PM, Ruediger Pluem <rpl...@apache.org> wrote: > > How about to require that the caller of ap_filter_setaside_brigade just hands > over a non NULL bucket_brigade as > buffered_bb (so changing apr_bucket_brigade ** to apr_bucket_brigade *) and > that it should handle *deferred_write_pool > as opaque structure and that it should not allocate from it?
Can that be enforced by the API as I proposed? > Or we request the caller to provide a non NULL > deferred_write_pool as well (so changing apr_pool_t ** to apr_pool_t *) and > warn the caller that the pool might be > cleared during the call of ap_filter_setaside_brigade. Hence solving all > lifetime issues would be in the responsibility > of the caller. IMHO that would complicate the caller's job/code, probably doing what's in ap_filter_setaside_brigade() now (use a subpool to avoid the leak, manage the cleanup..). The function would become a simple ap_save_brigade() call (and data_in_output_filter++), no real gain... Regards, Yann.