Thanks a lot for the suggestions. What are "streaming bucket types"?
For the buckets that nonblocking read returns success and zero, I tried
blocking read again and it still returns success and zero length, so I
just pass it to process the next bucket. It seems to work fine. Not
sure why there are zero length heap buckets.
I have also seen cases where apr_bucket_read() for heap bucket returns
a length larger than 8192, e.g. 8271, and processing of this buffer
caused a segmentation fault due to address out of bound. Any idea why
this could happen? I thought all bucket has a size less than 8192.
BTW, your apache book has been very helpful to me. Thanks for the good
Nick Kew wrote:
On Sat, 12 Jan 2008 19:18:10 -0500
Christina Fu <email@example.com> wrote:
I am writing a connection level input and output filter. I have
confusions about whether to use APR_NONBLOCK_READ or APR_BLOCK_READ
when calling apr_bucket_read() function. Could someone advise the
guidelines of using blocking vs non-blocking bucket read?
There's nothing specific to Apache. Blocking reads are easier to
work with, but preclude certain forms of optimisation.
I have seen situations when I use APR_NONBLOCK_READ to read a heap
bucket, it returns status of APR_SUCCESS but length is zero.
Sounds unlikely, unless the bucket had length zero. It should
only make a difference on streaming bucket types.
case, what shall I do with this bucket? shall I read it again with
blocking? or pass it and process the next bucket?
If you've nothing better to do, then yes. A filter function can
also return to its caller if you don't want blocking.