Building agents in Pharo

Hi Andy,

> On 21 Dec 2014, at 20:00, Andy Burnett <[hidden email]> wrote:
>
> I am interested in building a system that uses co-operating agents to parse events.
>
> Is Pharo a suitable system for this? I ask because given that it is single threaded, I wondered whether running multiple agents would present any sort of problem. Has anyone got experience of building something like this? And, if so, what did you use for your basic queue? Is there a suitable FIFO style queue within Pharo?
>
> Cheers
> Andy

The standard VM is using a single OS process (and thus uses a single core) but manages its own user land threads called Processes. The cool thing is that you can read/understand all this threading code (ProcessorScheduler, Process, Semaphore, Monitor, Delay, ..). There few if any limits to the amount of processes you can have.

To answer the second question: I believe SharedQueue is the class you should be looking at (first).

Sven

Andy which kind of agents would you like to build?
I was playing with Actalk (I should resurrect it in Pharo 40.
But in essence an object with a thread and a mailbox

Stef

Sven wrote
<<
The standard VM is using a single OS process (and thus uses a single core) but manages its own user land threads called Processes. The cool thing is that you can read/understand all this threading code (ProcessorScheduler, Process, Semaphore, Monitor, Delay, ..). There few if any limits to the amount of processes you can have. 

To answer the second question: I believe SharedQueue is the class you should be looking at (first). 

>>

Great, that sounds very encouraging. Thanks very much.

Andy

Hi Stef,

In very simple terms I want to build a system that would look at what you are typing in a text box, and have various agents that analyse the text and modify/augment it.

Simple Examples:

1. If I enter a web address it could look it up on Delicious.com and extract links that had been similarly tagged.
2. If I enter something that looks like a name, it would search google scholar, or mendeley and include links to any documents
   
3. If I entered "call david tomorrow" it would be able to add something to my things to do list.

My - very early - thoughts are that I need to continually parse the text, as I type, and various agents would look for the things they are interested in.  The question is how best to model this, and process the data.

You see this is what the color higlighting is doing.
You type and in parallel it is trying to colorize the text.
After I'm not sure that you need agents running all the time (reacting to each characters you type).
I would be you I would
 - prototype a hierarchy of classes whose instances have a kind of patterns to know if they should be kick in
 - at the end of each line, I would send the line to a list of such "agent" and react
  
- then only then, I would consider to fork the behavior of the agents. why? because when you can avoid concurrency this is simpler :)

Cheers

Andy