> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
> This was just committed to apr_util's bucket code.  This should fix the
> mod_include problem (at least it did for me).  What was happening, was
> that we read from the file, and converted it into a heap bucket.  Then, we
> destroyed the file bucket (free'ing *s at the same time), and then we used
> s->start to create the second file bucket.

It occurred to me while I was at lunch that there's no need to destroy the
shared+file bucket at all if we're going to turn around and create a new one
anyway.  All we're missing is an apr_bucket struct to point to the
shared+file bucket data we already have constructed.  Doing it this way
removes about two frees and three mallocs and a bunch of unneeded pointer
arithmetic from each file_read() call on non-MMAP platforms.  The patch is
attached due to a line wrap munging.

--Cliff

Attachment: fileread.patch
Description: Binary data

Reply via email to