The re-serialization is a constant-time cost, so not significant. The
question of efficiency has to be measured against the cost of development,
maintenance and consistency/stability of API semantics (the public API is
formally defined/enforced via the IDL). This is not to say that I'm against
the optimization, just that implementing it from the start would be a
premature denormalization, as the layering approach allows for exsiting to
do most of the work.

e

-----Original Message-----
From: Libbitcoin [mailto:libbitcoin-bounces@lists.dyne.org] On Behalf Of
Amir Taaki
Sent: Thursday, November 06, 2014 4:24 PM
To: libbitcoin@???
Subject: Re: [Libbitcoin] Direct websockets server

I don't mind that, the benefit though to direct native websockets
integration is less copying and re-serialising which is more efficient.
The intermediate step isn't necessary when we have the libbitcoin API
directly.

The Python gateway right now is very crappy since it's essentially
single-threaded blocking translator (in Python). It stops working often.
Something direct and integrated (in C++) would be more efficient and stable.

On 11/07/2014 12:18 AM, Eric Voskuil wrote:
> Yes, gotta provide web-friendly protocol and serializations!
>
> We can easily layer web sockets over ZMQ. That gets us into the browser.
> websocketpp sounds good, but I haven't looked closely.
>
> The next question is wire protocol. Using a client-side js lib the
> protobuf messages can be made accessible. ProtoBuf.js looks good there,
> but again haven't looked closely.
>
> We also have the option to map the wire protocol on the server to JSON,
> between websockets and ZMQ. This is also relatively straightforward -
> see pb2json.
>
> The harder part is getting the server up under the new/privacy protocol
> (libbitticoin-protocol). However that task is under way.
>
> In the mean time someone could take this up over obelisk
> (current/updated protocol). That just means leaving out the protobuf
> step, which means manually mapping the ZMQ messages to JSON behind
> websocketpp. Seems like a good path for the DarkWallet folks.
>
> e
>
> On 11/06/2014 11:30 AM, William Swanson wrote:
>> Yes, we definitely need websockets!
>>
>> This last weekend, I went to the Bay Area to help mentor a Hackathon.
>> The hackathon had over a hundred students, and they produced about 14
>> projects. Unfortunately, every single project, without exception,
>> relied on a centralized blockchain API like CoinBase or
>> Blockchain.info. This is not good! Companies like CoinBase are rapidly
>> turning into the new banking establishment, destroying the
>> decentralization we have worked so hard to achieve.
>>
>> If we want to reverse this trend, we need some easy-to-use,
>> web-compatible Blockchain data server to compete with these
>> proprietary services. The libbitcoin-node project can certainly
>> provide this, but not over ZeroMQ. Websockets are our best hope.
>>
>> -William
>>
>> On Thu, Nov 6, 2014 at 11:11 AM, Amir Taaki <genjix@???> wrote:
>>> I've been playing today with websocketpp, which only depends on boost
>>> and is still actively developed.
>>> It's simple to use and asynchronous (based off boost ASIO) which makes
>>> it ideal.
>>>
>>> Here's what I used on Ubuntu 13.10 to compile it:
>>>
>>> export BOOST_INCLUDES=/usr/include/
>>> export BOOST_LIBS=/usr/lib/x86_64-linux-gnu/
>>> scons
>>> cd build/release/echo_server/
>>> ./echo_server
>>>
>>> In another terminal window:
>>>
>>> sudo pip install websocket-client
>>>
>>> Create test.py and run it:
>>>
>>> from websocket import create_connection
>>> ws = create_connection("ws://localhost:9002/")
>>> print "Sending 'Hello, World'..."
>>> ws.send("Hello, World")
>>> print "Sent"
>>> print "Reeiving..."
>>> result = ws.recv()
>>> print "Received '%s'" % result
>>> ws.close()
>>>
>>> Other options:
>>>
>>> * QtWebSockets - not sure where it's used or how well developed it is.
>>> * libwebsockets - "lightweight pure C library built to use minimal CPU
>>> and memory resources, and provide fast throughput in both directions."
>>>
>>> libwebsockets seems very cool, especially the zero-copy write but it
>>> looks more complex to use:
>>>
http://git.libwebsockets.org/cgi-bin/cgit/libwebsockets/tree/test-server/tes
t-echo.c
>>>
>>> here's websocketpp for comparison:
>>>
https://github.com/zaphoyd/websocketpp/blob/master/examples/echo_server/echo
_server.cpp
>>>
>>> websocketpp should be easy to natively integrate as a transport into
>>> libbitcoin-server.
>>>
>>> _______________________________________________
>>> Libbitcoin mailing list
>>> Libbitcoin@???
>>> https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/libbitcoin
>> _______________________________________________
>> Libbitcoin mailing list
>> Libbitcoin@???
>> https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/libbitcoin
>>
>
_______________________________________________
Libbitcoin mailing list
Libbitcoin@???
https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/libbitcoin