the problem

End-user applications are huge, with tons of libraries used and are a pain in the ass to maintain. Libraries are huge, with a lot of things happening, with changes that break everything on a regular basis, with almost each time very few people working on it.

how to change that ?

Network protocols

Network protocols are awesome: very few changes, well documented, programming language agnostics.

Don't (just) write libraries, write applications !

Your need a functionality in your application, why do you have to code it ? Just ask a service !

Example

You want to download a file, you will always have the same input: a string corresponding to the file to get, such as ftp://example.com/file.txt. You don't have to worry about the protocol to use in your own application, the burden is on the dedicated downloading service.

benefits

Awesome abstractions.

You will be able to do things without any code.

• applications don't have to know if the services they use is on the network or on your own computer
• applications don't need to change anything to handle new protocols, no recompilation
• applications can be statically compiled, the memory footprint should be extremely low (yes, even for a browser)

Simple applications.

You only need to code the specific code of your own application. The only thing your application should have to take care is its functionality, and to communicate to services providing abstractions.

Consistency.

Everything will be developed in the same repository: same coding standards, changes will be tested on every provided applications…

Code review.

We should always try to provide new abstractions, reducing the code needed in both services and end-user applications. To that end, code review is a must.

Application and services

• Services: daemons providing a feature (windowing, audio, network, input, pubsub, …)
• Applications: end-user applications (browser, mail user agent, instant messaging app, …)

examples

A browser that can download everything, via every existing protocol. No any specific code in the browser itself and no configuration.

You want to play a game on your rasberry pi, but it is not powerful enough, run your application on your laptop but take the inputs from the rpi (or anywhere on the network) ! No specific code needed.

TODO

Figures, a lot of them, to explain everything.

connection init (draft)

how things happen

1. Service creates a unix socket /tmp/service-index-version.sock
2. Application connects to /tmp/service-index-version.sock

legend: - service: service name - index: process index (to launch a service several times) - version: service version

pure "networking" view (what should go in the sockets)

connection

1. Application connects to /tmp/service-index-version.sock
2. Service acknowledges (empty message)

disconnection

1. Application sends a message "CLOSE" to the server

data

1. Application or server sends a message "DATA", no acknowledgement

messages format

In order to communicate between the application and the service, we use the Type-Length-Value format. This will be used with some conventions.

programming, debug

overview

The format will be "type : value".

The type will be a simple byte :

* <0 - 15>   : control, meta data
* <16 - 127> : later use
* <128 - 255> : application specific (windowing system, audio system, …)

index   | abbreviation  | semantic
0       | close         | to close the communication between the application and the service
1       | connection    | to connect to the service
2       | error         | to send an error message
3       | ack           | to send an acknowledgment
4       | message       | to send data