2017-06-07 16:45 GMT+02:00 Esteban Lorenzano estebanlm@gmail.com:

On 7 Jun 2017, at 15:15, Nicolas Cellier <nicolas.cellier.aka.nice@gmai l.com> wrote:

2017-06-07 14:32 GMT+02:00 Esteban Lorenzano estebanlm@gmail.com:

...

Hi Dave, list

I want to commit some changes to OSProcessPlugin. Can I have rights to that repository?

thanks, Esteban

ps: I need to insist in the necessity of having all VM plugins (the “official” ones: those who are compiled with the osvm builds) in the same place. Otherwise this is a horrible (and irreproducible) process.

Hi Esteban, why not reproducible? Isn't it manageable with a Monticello Configuration Map (.mcm) as David suggested, or a Metacello ConfigurationOf... for being more pharo-to-date?

because is a pain to maintain, and having different repository services makes you needing to thrust in different providers status, and then if you want to work offline you need to take a copy of each repository which are dispersed all around the world.

The location are relatively stable so far (one move from www.squeaksource.com http://www.squeacksource.com to source.squeak.org for VMMaker a few years ago).

I just lost 3 hours of my life trying to figure out which version of OSProcessPlugin was the correct one, then looking for the correct repository, etc. This wouldn’t happened if all used plugins are maintained together.

... Nor if you would use either a mcm or Metacello configuration.

Speaking of repository map, I just checked the status of Pharo 6, and it does not look more simple:

[image: Images intégrées 1]

Naming branches and tagging stability is a different problem than repository location.

For OSProcess, there are only two branches (interpreter and oscog/spur), and history is quite linear (just bug fixes). So the policy is: load latest, run regression tests, revert only if something goes wrong.

It's true that in MC, explicitely naming branches is optional, but we can... It's just a convention (like naming master trunk cog or anything else is just a convention in git). I'm not sure we really need it yet for plugins, new features are scarce.

The main difference is not about branches, it's more about integration of tools, issues/wiki/PR/CI.

btw (advocating for the way I would like to have things) this wouldn’t even be necessary if all packages needed would be versioned along the platform sources, using filetree format… then having the clone of your desired version you would be sure to have also the correct VMMaker AND plugin versions… but I know this way of sorting things is too much to ask, so I would be happy if all packages are *at least* in the same place.

Esteban

For having worked a bit on the pharo-vm two years ago, the workflow was horrible.

We had the workspace provided by image state - which is like a mixture of working copy (dirty with unpublished changes) + staging area (cherry picking thru MC GUI), eventually with branches/stach saved in package-cache, or in other MC repositories.

This was concurrencing the working copy on disk, which was often out of sync, and conflicting.

Branches in MC repositories had to be synced manually with git branches (I used my own Smalltalkhub repository for my own experiments with many feature branches, plus official VMMaker repo which I regularly tried to merge, plus pharo VM github repo).

Each pull request was rotting immediately due to MC metadata and incomprehensible policy of ignoring the merge tool provided by Thierry Goubier (which was working quite smoothly). And there were bugs introduced by pharo in MC, plus bugs of file names too long, and I probably forget other annoyances.

So it was a nice experiment for learning, but clearly not mature.

Shifting entirely to git would reduce the complications a bit. and I promise I will test the new git-based-in-image-version-control-tools provided by Pharo and help getting closer to your ideal world. But what about the status of image changes? Or is the image a pseudo file system? Anyway, you know it will take time to reach the maturity and smoothness of MC. MC rocks (at least in Squeak).

Stability of tools apart, for OSProcessPlugin, we have a different case: the manager wants to control integration of change requests. How is it going to work with a single repository? Or is it something you want to change too?

2017-06-07 22:06 GMT+02:00 Esteban Lorenzano estebanlm@gmail.com:

forget it, I renounce to try to explain why from maintainability/management point of view is a lot better to keep all sources together. Obviously, I’m wrong, and the situation we have today is ideal.

Esteban

Hey Esteban, You have the right to disagree, my arguments are not gospell truth, and I didn't remember writing ideal anywhere (just MC rocks).

IMO, having VMMaker sources replicated on pharo-vm github repo thru filetree does count for something in managing complications you have so far.

I said it was not sustainable two years ago, and I doubt it's more now.

For the long term goal, I did not say that shifting to 100% git/github tools would not be sustainable. I just said that such switching does require mature in-image-tools, or guys like Eliot and me would not switch, take it for some sort of wisdom.

For the short term goal, the original question of moving OSProcessPlugin, I did the mistake (I missed the latest version recently), Eliot did the mistake (he has even overriden one fix of mine by using an older revision end of last year), you did the mistake, so it clearly shows something about the process.

This process relies too much on manually opening repositories, inspecting changes, and deciding to load/merge a newer MC revision.

Putting all the packages under one repository would ease this manual process. The question I ask is whether there exist other means for integrating changes as long as they are correctly managed upstream.

Eliot suggested that, David should generate C source and publish on opensmalltalk-vm soon after posting a production ready MC package: indeed, since he his the integrator, he knows what is production ready. The danger is to forget, or postpone, especially because doing it both for interpreter and cog/spur is too much demanding (too many different tools maybe?). One more step, and we go to automation of C-cde generation :), we just need some sort of human blessing from the integrator.

David suggested that squeak guys use mcm, and I completed with Metacello. You could have answered that you, as integrator of Pharo VM, wants finer control than automated update to MC head if you wanted to argue, but having to argue for you is no fun ;)

Last thing, such decision cannot be made without asking the author/maintainer of the plugin, if you want to convince him, you have to argue too, no?

Best regards

Nicolas

On 7 Jun 2017, at 21:43, Nicolas Cellier <nicolas.cellier.aka.nice@ gmail.com> wrote:

2017-06-07 16:45 GMT+02:00 Esteban Lorenzano estebanlm@gmail.com:

...

On 7 Jun 2017, at 15:15, Nicolas Cellier <nicolas.cellier.aka.nice@gmai l.com> wrote:

2017-06-07 14:32 GMT+02:00 Esteban Lorenzano estebanlm@gmail.com:

...

Hi Dave, list

I want to commit some changes to OSProcessPlugin. Can I have rights to that repository?

thanks, Esteban

ps: I need to insist in the necessity of having all VM plugins (the “official” ones: those who are compiled with the osvm builds) in the same place. Otherwise this is a horrible (and irreproducible) process.

Hi Esteban, why not reproducible? Isn't it manageable with a Monticello Configuration Map (.mcm) as David suggested, or a Metacello ConfigurationOf... for being more pharo-to-date?

because is a pain to maintain, and having different repository services makes you needing to thrust in different providers status, and then if you want to work offline you need to take a copy of each repository which are dispersed all around the world.

The location are relatively stable so far (one move from www.squeaksource.com http://www.squeacksource.com/ to source.squeak.org for VMMaker a few years ago).

...

I just lost 3 hours of my life trying to figure out which version of OSProcessPlugin was the correct one, then looking for the correct repository, etc. This wouldn’t happened if all used plugins are maintained together.

... Nor if you would use either a mcm or Metacello configuration.

Speaking of repository map, I just checked the status of Pharo 6, and it does not look more simple:

<Capture d’écran 2017-06-07 à 20.36.36.png>

Naming branches and tagging stability is a different problem than repository location.

For OSProcess, there are only two branches (interpreter and oscog/spur), and history is quite linear (just bug fixes). So the policy is: load latest, run regression tests, revert only if something goes wrong.

It's true that in MC, explicitely naming branches is optional, but we can... It's just a convention (like naming master trunk cog or anything else is just a convention in git). I'm not sure we really need it yet for plugins, new features are scarce.

The main difference is not about branches, it's more about integration of tools, issues/wiki/PR/CI.

...

btw (advocating for the way I would like to have things) this wouldn’t even be necessary if all packages needed would be versioned along the platform sources, using filetree format… then having the clone of your desired version you would be sure to have also the correct VMMaker AND plugin versions… but I know this way of sorting things is too much to ask, so I would be happy if all packages are *at least* in the same place.

Esteban

For having worked a bit on the pharo-vm two years ago, the workflow was horrible.

We had the workspace provided by image state - which is like a mixture of working copy (dirty with unpublished changes) + staging area (cherry picking thru MC GUI), eventually with branches/stach saved in package-cache, or in other MC repositories.

This was concurrencing the working copy on disk, which was often out of sync, and conflicting.

Branches in MC repositories had to be synced manually with git branches (I used my own Smalltalkhub repository for my own experiments with many feature branches, plus official VMMaker repo which I regularly tried to merge, plus pharo VM github repo).

Each pull request was rotting immediately due to MC metadata and incomprehensible policy of ignoring the merge tool provided by Thierry Goubier (which was working quite smoothly). And there were bugs introduced by pharo in MC, plus bugs of file names too long, and I probably forget other annoyances.

So it was a nice experiment for learning, but clearly not mature.

Shifting entirely to git would reduce the complications a bit. and I promise I will test the new git-based-in-image-version-control-tools provided by Pharo and help getting closer to your ideal world. But what about the status of image changes? Or is the image a pseudo file system? Anyway, you know it will take time to reach the maturity and smoothness of MC. MC rocks (at least in Squeak).

Stability of tools apart, for OSProcessPlugin, we have a different case: the manager wants to control integration of change requests. How is it going to work with a single repository? Or is it something you want to change too?