On Tue, Nov 25, 2008 at 08:35:29AM +0100, Ruediger Pluem wrote: > On 11/25/2008 02:26 AM, [EMAIL PROTECTED] wrote: > > URL: http://svn.apache.org/viewvc?rev=720374&view=rev > > Log: > > Clarify the behaviour of the apr_bucket_read() and apr_brigade_write() > > functions to lift some of the mystery surrounding how they work. ... > > --- apr/apr-util/trunk/include/apr_buckets.h (original) > > +++ apr/apr-util/trunk/include/apr_buckets.h Mon Nov 24 17:26:46 2008 > > @@ -800,6 +800,20 @@ > > > + * > > + * Some buckets, such as the file bucket, might have a fixed size, > > + * but be significantly larger than is practical to store in RAM in > > + * one go. As with the socket bucket, if an attempt is made to read > > + * from a file bucket, the file bucket is magically morphed into a > > + * heap bucket containing a convenient amount of data read from the > > IMHO this is not correct. It depends on whether MMAP is available and > allowed. If MMAP is available and allowed the file bucket is morphed > into an MMAP bucket of reasonable size.
Right, but none of this should be part of the API contract. The specific behaviour of any bucket type on _read() is an implementation detail. joe
