Re: [GNUnet-developers] GSoC

[Martin Schanzenbach]

Hi,

On Sat, 2016-02-27 at 12:39 +0100, carlo von lynX wrote:
> Hello again, Dan & Martin!
> 
> On Fri, Feb 26, 2016 at 04:48:10PM +0200, Daniel Golle wrote:
> > 
> > > 
> > > Great. We already have an extensive jspsyc library that we used
> > > in
> > > the psyczilla add-on for mozilla. It uses native sockets.
> > That's the kinda stuff I was thinking of. JSON is useful beyond
> > HTTP.
> Actually JSON is less efficient than PSYC, so we don't use it.
> See http://www.psyc.eu/libpsyc/bench/benchmark.html
> 
> If it'S true that with dbus my computer is parsing and rendering
> XML all the time in a sort of digital masturbation, and you tell
> me that ubus does the same using JSON, well then I am sorry the
> world hasn't developed psycbus instead.
> 
> Picking up on the thread you are having with Martin, I understand
> the power of ease of use.. but I haven't understood the power of
> REST in our use case. How can we leverage HTTP features like
> caching if our architecture is mostly about push and there are
> no query/cache/retrieve patterns? I read the stackoverflow debate
> and still don't understand how it works for us.
> 
Yup. I think I am trying to solve a very different, a lot more
pragmatic problem:
Imagine you want to write a nice GUI for the filesharing part of GNUnet
that communicates with a few service over the API.
Now imagine your GNUnet peer is on a Raspberry PI / "In the cloud" / On
your desktop PC and you want to use that peer from your laptop/mobile
phone using the GUI.
At this point, if your GNUnet node does not expose the API over REST
you are welcome to use sockets/custom JSON-RPC in your code for mobile
(Android) and browser (JS) but then say hello to the year 1998.
Today, apps communicate with services through rest. Developers are used
to this and it does make sense because ALL platforms can handle it
without additional software stacks.

- Martin

> > 
> > > 
> > > Huh! Where did you read that? Would I suggest to use PSYC as a
> > > mail
> > > system if this was true?
> > I suppose not ;) And that made me wonder...
> Mybe it'S because of the historic meaning of the acronym.. which is
> a popular problem.. like HTTP is only for hypertext.. SMTP is indeed 
> simple and SSL is secure. I bet there are more hilarious cases of
> things
> named after goals they have outlived in a good or bad way.
> 
> > 
> > Maybe I just remember it worngly, but the last time I touched PSYC,
> > messages were either delivered right away or lost. Logging-in after
> > a connection failure would not show me the history of messages I
> > had
> > missed, but rather just what ever happened from then on.
> There is a drastic change in architecture between the PSYC1 designed
> in 1994 and the PSYC2 that runs over GNUnet. You are probably
> thinking
> of whenever you used a psyced server which can only store messages
> for you if you register your account on it - otherwise it doesn't
> know who 
> it is keeping messages for and anyobody else could step in and read
> them.
> 
> Or maybe you just tried to use it with a Jabber client. psyced has a
> bug for Jabber clients whereby stored messages aren't always
> delivered
> successfully (XMPP is only fully supported for interserver).
> 
> Anyway, the problems of an old PSYC1 server are irrelevant in the
> new serverless architecture.
>