On Mon, Jul 18, 2011 at 4:19 PM, Cliffor= d Heath <c= lifford.heath@gmail.com> wrote:

Awesome stuff, Yehuda! I look forward to this making its way into a distrib= ution.

On 18/07/2011, at 4:44 PM, Yehuda Katz wrote:

=A0 =A0 =A0 =A0=95 It is possible to read_nonblock from Net::HTTP response= , even if the response has Transfer-Encoding: Chunked, keepalive or compres= sion. This makes it possible to use Net::HTTP in reactor libraries, instead= of writing whole new HTTP libraries for each kind of reactor

Does this also work with SSL?

Yes, assu= ming that the SSLSocket is, in fact, as polymorphic with Socket as it claim= s to be.

=A0

I recall seeing tweets from you questioning why reading from an SSL socket<= br> might be affected by whether the socket is currently writable - I assume yo= u
worked out that it's due to handshaking (which can be re-initiated at a= ny point
during an SSL connection). Even though the application layer is in a mode where it expects only to read, the handshake may require writing, and hence=
can block on write if the socket is currently stalled.

Yep! I worked it out. Thanks for the full explanation :)

=A0

=A0 =A0 =A0 =A0=95 Make the body parsers (regular body and chunked encodin= g) separate objects that can support either blocking read or non-blocking r= ead.

Heh. Now all we need is a revision of the Rack API to allow reading chunked=
bodies in a piecemeal fashion. They can already be sent, because the
response body may be enumerable - though that's still not reactor-frien= dly.

I use async_sinatra which hacks Thin to make the EventMachine reactor
work correctly - see http://github.com/raggi's projects.

Integration to support websockets is also high on the list of wants. You sh= ould
ask on <mailto:socket_io@googlegroups.com> how best to do that, since= many
of the interested parties hang out there.

=A0 =A0 =A0 =A0=95 Improving the tests to test non-blocking read across al= l kinds of requests, and add tests for decompression

I'd be interested in how you test non-blocking for SSL connections, see= ing as
you need a custom SSL server which triggers renewed handshakes to do it
properly...?

I'm currently simply u= sing the tests that shipped with Net::HTTP in the current stdlib, but I'= ;d be very interested in ways to improve them. Unfortunately, the current S= SL tests are not as exhaustive as the other tests (such as the ones that te= st gzip or Transfer-Encoding: chunked), but I would like to improve that.

=A0

Clifford Heath, Data Constellation, http://dataconstellation.com
Agile Information Management and Design
Skype: cjheath, Ph: (+61/0)401-533-540

--20cf307f378278b11d04a852f81f--