I mentions this because in OSProcess-Unix there is UnixProcess class, at the class side there are several examples that launch a separate instance of Squeak/Pharo and make the two communicate with each other. In the instance of the class there are methods like forkHeadlessSqueak , forkHeadlessSqueakAndDo: etc. So the functionality is there but since Mariano said he will be working to improve OSProcess I thought that would a very useful area to work on. Since not only will allows running on multiple cores but even to lift any memory limits. Also I think this is a nice way to make the pharo enviroment which I mean the running vm with an image and satelite files not only a collection of objects but also a object itself that can communicate with other pharo enviroment instances (Pharo running processes) . Fuel I assume enters here the equation as a data exchange format , the problem I have with fuel is that its not backward compatible which for me is a very big deal especially when we talk about threading and processes which is a pretty fundamental feature. We need something that will be stable but also code orientated that will remain backward compatible because the last thing you want is initiating multiple processes that cant understand each other because they happen to use diffirent image versions that use diffirent verrsions of Fuel or what else. On Sun, Dec 20, 2015 at 1:46 AM David T. Lewis <lewis@mail.msen.com> wrote:

Mariano,

I am really happy to see this and I look forword to helping you in any way I can.

Dimitris,

It is really interesting that you should suggest this. I think it may be a bit off topic for Mariano's initial goals, but Mariano is the driving force behind Fuel, which is the serializer that I use for RemoteTask in OSProcess/CommandShell. This attempts to do much of what you suggest, using Fuel to return the results of processing executed in a forked image (in the unix sense of "forked", i.e. extremely memory efficient).

So maybe if I can help Mariano with his project (which of course I plan to do anyway), then he might be able to help me turn RemoteTask into something reliable enough for general purpose use.

Another nice connection here is that Eliot's Spur image model will make #forkSqueak (the primitive behind RemoteTask) more efficient in cases where the cooperating images do a lot of processing, because memory changes are more localized in the Spur memory model. In principle it should be possible to run very large object memories in 64-bit Spur, while forking cooperating images that consume very little additional real memory.

Dave

On Sat, Dec 19, 2015 at 01:33:07PM +0000, Dimitris Chloupis wrote:

...

I am very excited for such a project.

I hope you manage to fix the stdout problem on macos.

One thing I would like to see is an expansion of the ability of OSProcess to launch multiple instances of pharo and make them communicate with each other. Thats a way to have pharo running on multiple cores. Maybe establish a framework of communication of images that will be based on the existing design of the fork mechanism in Pharo.

Python is doing something similar with multiprocessing module which launches multiple processes of cpython executable and makes them communication with each other through a similar api to its threading module. This way python takes advantage of multiple cores. Python threads like pharo are green threads and pharo like python has a GIL like mechanism.

Taking into account that current computers have at least 4 cores, its not a bad idea to speed up pharo like this at least 400% ;)

On Sat, Dec 19, 2015 at 3:24 PM Mariano Martinez Peck < marianopeck@gmail.com> wrote:

...

Dear all,

I am tremendously happy to announce you that Pharo Consortium will sponsor yet another development effort. In this particular case, it's my honor to carry on such an effort and I will be developing and improving a few things and ideas. All the contract and paperwork has already been done so it's time to start working.

Regarding the developments itself, we discussed about these topics:

1) Experiment with a simpler yet more limited OSProcess alternative to execute OS commands. 2) Improving FileSystem in order to better deal with some POSIX stuff like symbolic links, unix file permissions, etc etc.

I have already started with 1). OSProcess is super complete and it involves some packages (OSProcess / CommandShell) as well as some VM plugins (OSProcessPlugin, AIOPlugin). OSProcess provides lots of features (like forking the running image) but we would like to focus only in executing OS commands. In addition, OSProcess dates from quite some years ago when many of the current infrastructure features did not exist yet.

So the idea is to think a simpler alternative to execute OS commands, using new features such as threaded FFI (for example, for reading async from pipes), pinned objects, etc. We want to use FFI as much as possible rather than VM plugins.

From the image side, we are thinking about an API which would look like a builder-like API (FileSystem, XStream, etc).

We have been discussing with David Lewis (OSProcess author) as well as with Eliot Miranda, Esteban Lorenzano, Damien Pollet, Stephane Ducasse, etc in order to agree in a project that would be worth doing (otherwise we would continue using existing tools). And they have all been very kind with me providing positive discussions.

I will soon do a survey to measure the most common use cases of OSProcess and related tools.

If you have any thought, question, feedback, or whatever, please let us know.

Finally, note that I will be doing this project together with another client's project (Quuve) and my life does not have much free time these days...so please be patient!!!

Best,

-- Mariano http://marianopeck.wordpress.com

not really because as I said UnixProcess already has an implementation for doing this, you dont need any of those protocols/libraries you mentioned, it does not even need Fuel. It already works. And the reason why I mentioned Python multiprocessing module is because multiprocessing does not only implement a specific way of communication between processes its also handle concurrency issues like data synchronization and much the of messy problems that come with concurrency. And this is the are of improvement, its not what kind of format the communication is , as you said it should be up to yourself how to format your date, maybe you want json, maybe fuel, maybe you have your object and classes. The real focus here is true concurrency issues on may experience because you will be having two image running alongside with no guarantess when one finish on action and framework to synchronize data. Obviously one can implement his own but I think Pharo would greatly benefit to have an establish default framework , even a small one for OS-Process concurency. On Sun, Dec 20, 2015 at 11:09 AM Pierce Ng <pierce@samadhiweb.com> wrote:

On Sun, Dec 20, 2015 at 08:56:43AM +0000, Dimitris Chloupis wrote:

...

Fuel I assume enters here the equation as a data exchange format , the problem I have with fuel is that its not backward compatible which for me

Once you have a set of OS processes running Pharo how they talk to each other is up to you no? E.g. the Pharo instances could use ZeroMQ, XML-RPC, JSONRPC, msgpack, coordinate via Redis/SQLite/PostgreSQL, etc.

Pierce