[kepler-dev] starting an Actor IO package

[Tristan King]

Hi Paul, replies below.

On Fri, Jul 4, 2008 at 4:53 AM, Paul Allen <pea1 at cornell.edu> wrote:

>  Tristan, I think that this is the beginning of a valuable direction to
> head with Kepler (and ptII). I have a few questions and thoughts about this,
> but bear in mind that I haven't actually grabbed your prototype code to look
> at it.
>
> 1) For the "Input" side of things, are you thinking that the IO package
> would assist actors interacting with uses (e.g., picking the image format
> type "png" from the pulldown menu in the "ConvertImage" actor configuration)
> . Can anybody suggest some actors to play with that have a rich user
> interaction? I haven't really come across any. What are the use cases
> involved with them?
>
If you mean the options of the actors before the workflow is started, then
no. For input, what I mean is any user input that happens when the workflow
is executing. I know there're an actor or two out there that pops up a
dialog and waits for user input, but I couldn't point them out myself,
anyone else?  For output, it is simply the visualisation of the results,
i.e. the display window, or any graphs produced, that sort of thing.

>
>
> 2a) Or, or are you thinking that the IO package primarily deals with
> abstracting the file system that many actors depend upon?
>
I think that an abstraction of the file system would be quite handy. One of
the things I have to deal with in hydrant is all the actors that write and
read files and making sure they don't read anything they're not supposed to
and that they write to a specific directory. If these actors went through a
file system abstraction I would only have to preform these checks in the
"device" not in every actor that uses them.

>
>
> 2b)  Even with a solid file system (or workspace) abstraction, I think
> there is still a need/use for a separate, flexible approach that actor
> authors can use to provide output to the user.
>
> Tristan King wrote:
>
> I've started doing a bit of work on building an actor IO package (
> ptolemy.actor.io)
>
> The purpose of this is to provide user input and output for actors so that
> the actors themselves don't have to be responsible for how they get user
> input, or how output is displayed. This will help to build actors that are
> useable in gui and headless environments without having to have modified
> actors for each situation.
>
> I've built a simple prototype which uses an IO factory which is embedded in
> the Manager (since all actors can get access to the manager which controls
> them). When an actor requires an IO "device" it makes a request to the IO
> factory which returns an instance of the desired "device". For example, the
> Display actor requests a Text-based output device. This device implements a
> standard text based output device interface which has a write(String)
> function, and all the Display actor has to do is "write" the text to the
> device which handles how to display it. I've used spring to instantiate the
> correct IOFactory so that in the case of vergil, each manager is loaded with
> a ptolemy.vergil.io.VergilIOFactory and if vergil is not used the manager
> defaults to a ptolemy.actor.io.HeadlessIOFactory.
>
> My work is currently stored in my ptII git mirror, in a branch which has
> been converted for maven compilation.
> If you want to clone the whole repository here are the git commands to
> clone the repository and to get to the branch (this will take a while):
>
> git clone git://git.hpc.jcu.edu.au/jc124742/ptII.git
> git checkout -b actorio origin/actorio
>
> or you can download the current snapshot of the branch (which is about
> 35mb) from:
>
> http://www.hpc.jcu.edu.au/git/?p=jc124742/ptII.git;a=shortlog;h=refs/heads/actorio
> you can then run it by doing:
>
> mvn compile # compiles the code, may take a while the first time while
> maven sorts out it's dependencies
> mvn exec:exec # starts vergil
>
> then fire up any workflow while uses a Display actor. Note that i'm rubbish
> with swing and some of the window placing/sizing functionality has been
> removed to facilitate the separation of the actor from the visualisation, so
> the display window isn't nearly as good as the old one. If you start the
> workflow using MoMLSimpleApplication (i.e:
>
> mvn compile
> java -cp
> target/classes:$HOME/.m2/repository/org/springframework/spring-context/2.0.8/spring-context-2.0.8.jar:$HOME/.m2/repository/org/springframework/spring-beans/2.0.8/spring-beans-2.0.8.jar:$HOME/.m2/repository/org/springframework/spring-core/2.0.8/spring-core-2.0.8.jar:$HOME/.m2/repository/log4j/log4j/1.2.15/log4j-1.2.15.jar:$HOME/.m2/repository/commons-logging/commons-logging/1.1.1/commons-logging-1.1.1.jar
> ptolemy.actor.gui.MoMLSimpleApplication ../test.xml
>
> ) it will print the output of any Display actors to the console.
>
> If other people want to work on this, I will create a new branch for it on
> the ptolemy II svn (it requires a few changes to the core of ptolemy so i'm
> not gonna commit it to the trunk) and put it there in makefile building
> mode.
>
> So, is anyone else interested in this? I think it's a key component in
> making kepler available in multiple environments. But it's a big change,
> much too big to be left up to one person. I would feel a lot more
> comfortable with a bunch of people telling me my ideas and designs are
> rubbish and help me build a good system rather than to build something that
> no one but me wants to use.
>
> Things that need to be done:
> * My factory idea and prototype architecture needs to be criticised.
> * Types of "devices" need to be defined (i.e. text output, image output,
> file output, graph output, etc)
> * IO Factory and device interfaces need to be defined.
> * Existing actors need to be modified to use the factory.
> * standard factory implementations need to be written (i.e. I'm rubbish at
> swing and thus someone else should do the vergil factory).
>
> Comments? Ideas? Criticisms?
> Cheers
> -Tristan
>
> --
> Tristan King
> Research Officer,
> eResearch Centre
> James Cook University, Townsville Qld 4811
> Australia
>
> Phone: +61747816902
> E-mail: tristan.king at jcu.edu.au www: http://eresearch.jcu.edu.au
>
> ------------------------------
> _______________________________________________
> Kepler-dev mailing listKepler-dev at ecoinformatics.orghttp://mercury.nceas.ucsb.edu/ecoinformatics/mailman/listinfo/kepler-dev
>
>


-- 
Tristan King
Research Officer,
eResearch Centre
James Cook University, Townsville Qld 4811
Australia

Phone: +61747816902
E-mail: tristan.king at jcu.edu.au www: http://eresearch.jcu.edu.au
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mercury.nceas.ucsb.edu/ecoinformatics/pipermail/kepler-dev/attachments/20080704/212f869e/attachment.htm