Why doesn't Iceberg checkin other assets (scripts) but does check them out?

On 15 Jun 2018, at 10:29, Tim Mackinnon <[hidden email]> wrote:

In many parts of the dev world - every commit should be shippable, meaning atomic - particularly with the use of CI systems (Travis, Gitlab etc) that build on every commit. 

Of course you can choose to work a different way - but our tools shouldn’t force a very common practice to be awkward, should they?

I don’t understand why many of you guys are so anti such a common thing.

In fact our community pioneered this way of working - a build machine where you loaded all your code and ran all your tests before committing.  Also, a large number of agile teams will council to work on master (and avoid branches - a technique we also pioneered).

In today’s world where we live in a wider polyglot stack this is one of the reasons why git is so attractive as it should easily embraces that.

I get that it’s not quite possible yet - but there are some decent ways to accommodate this (a staging operation, or perhaps a better way to make use of the #addToIndex: which I will experiment with).

Anyway, I’ll shut up - and possibly will learn how to implement it myself. Thinking more, I may also be able run with commit and no push , which I will try - as it’s probably the easiest mechanism if the CI will recognise that (I think it will).

Thanks for the enlightening conversation, and as always the hard work on iceberg that made this conversation even possible.

Tim

Sent from my iPhone

On 15 Jun 2018, at 07:11, Norbert Hartl <[hidden email]> wrote:

Am 14.06.2018 um 13:12 schrieb Thierry Goubier <[hidden email]>:

Hi Norbert, Tim,

2018-06-14 11:33 GMT+02:00 Norbert Hartl <[hidden email]>:

Am 14.06.2018 um 10:30 schrieb Tim Mackinnon <[hidden email]>:

Hi - yes I’m pleased you check out the entire tree, although currently it’s a bit confusing that you do (fortunately this does give the possibility that we can checkout images and other resources that an Pharo application might rely on - without having to resort to the Seaside FileLibrary trick).

However my concrete case was that I have a gitlab ci pipeline and next to my src directory in my project I have a config directory that has some Nginx config for my teapot app. If I add a teapot route, I also need to adjust that config and check both changes in together. I can’t easily do that now?

I can modify /config/app.nginx either in another app (intellij) or even in the simple Pharo text editor, and the I can add my new route in my DemoApp>>createRoutes method but how do I check them in together so my pipeline will build atomically?

Iceberg hasn’t written out the changes yet, so IntelliJ can’t see them to do a commit, and iceberg ignores the parallel /config directory (that it checked out). So it’s a catch 22.

This is why I suggested maybe we could specify safer (textual) directories that iceberg might also checkin? OR we have a Stage command in iceberg that does everything that commit does up to the point of actually writing to the repo - then I could jump to IntelliJ and do the final commit there and use its tools to manage non Pharo stuff (until we can build more)?

Does this make sense?

I don’t understand why you are so eager to have everything into one commit. Usually the tension is rather have small commits. What is the problem of having two commits for this?

A single commit allow one to make sure that both the smalltalk code and the external resource are well in sync, and that you may not ending up in a situation were at commit j has data format v1 and smalltalk code for format v2, because it is in commit j+1 that you rewrote the data in format v2.

Yes, sure but that is only a problem if you lost control over which commit goes into action. I think the problem is in the deployment then. Putting integrity constraints on commits is IMHO the wrong level deployment granularity

Norbert

I had that issue recently with a Pharo / FPGA project with a Pharo package generating state machines to be integrated in a Verilog FPGA design with C code for the drivers, the softcore in the FPGA, and test programs, and both Pharo and the C/FPGA working out of the same test data... And that whole thing getting regularly out of sync.

IMHO: I'd model a concept of "multi-lingual projet" for that sort of things, where one can list, in Pharo, Monticello packages and external files or directories. Whatever the technology you use (OSProcess, libgit, whatever...) it is easy then to commit all this in a single pass. On a per-package basis, I'd see the package manifest as a suitable place for listing external resources. On a per-project basis, a possible place could be the manifest for the BaselineOf the project.

Thierry

 

Norbert

As an aside - I’d really like to checkin in the play-xxx directories (the .ph files) as there is often useful playground stuff I’d like to access on my home computer. We can’t do that easily at the moment either.

Tim

Sent from my iPhone

On 14 Jun 2018, at 09:12, Guillermo Polito <[hidden email]> wrote:

Just to complement Esteban's answer:

- Iceberg checks out in disk more than the src directory because you **may** want to edit files from the command line, and after long discussions we did not want to forbid that.

Actually, just to put everybody in perspective, at first the idea was to not have a working copy in disk at all, but just hit to the blob.

Imagine is nowadays we are a bit alien, that would have been worst :)

- About checking in files. I'd like to understand what you mean exactly.

  - Do you want to load them into memory?

    This would be the "more consistent" way to do it, following the "the image it its own working copy" metaphore.

    This would allow us to, for example, share an image and transparently share resources with it (without requiring to clone).

    But this would have some impact in memory consumption and add stress to the GC, right?

  - Or do you mean to ask like any other Git client and show you the file differences between the working copy and the git index?

    The problem with this approach is that we will have some treatment for pharo code and some different treatment for non-code...

    If I do a change to a class, the change is kept in the image. But if I do a change to a file, that change is not kept in the image!

Also, as Esteban says, having an IDE with support for files would mean that we would need good tools to edit in-memory files (not only text files, right? but also any kind of binary file...)

So far we cover the bare minimum that allows us to *not lose* changes :)

On Wed, Jun 13, 2018 at 11:07 PM Tim Mackinnon <[hidden email]> wrote:

> yeah… but is a lot of work and no time just right now.
> long term, it would be cool to manage everything from iceberg.
> but reality check, is a huge amount of work so it has to come step by step.

Fair enough - its pretty cool we’ve got this far, and I guess the onus is on the rest of us to learn more about how its done and see if we can contribute more somehow. I really appreciate the love you’ve already put into this - it works far better than I think we even realised it could.

Tim



> On 13 Jun 2018, at 21:55, Esteban Lorenzano <[hidden email]> wrote:
>
>
>
>> On 13 Jun 2018, at 22:44, Tim Mackinnon <[hidden email]> wrote:
>>
>> Esteban - so I don't then understand why iceberg (usefully in my view) checks out more than the src directory, if it’s only focusing on the Pharo blob?
>>
>> I’m guessing that by knowing where the src is, you are just committing that part of the tree with libgit?
>>
>> Perhaps from a pragmatic first step you might consider letting us add a second safe resources directory that you could check in atomically as well (on the understanding all bets are off if it goes wrong?)
>>
>> OR could we have a check in mode does all the add/remove operations and writes to disk but then let’s you drop to the command line/other tool to add any other files and do the final commit?
>>
>> I just feel like you/we are so close to something that works a bit more broadly and embrace the wider world.?
>
> yeah… but is a lot of work and no time just right now.
> long term, it would be cool to manage everything from iceberg.
> but reality check, is a huge amount of work so it has to come step by step.
>
>>
>> Tim
>>
>> Sent from my iPhone
>>
>>> On 13 Jun 2018, at 21:28, Esteban Lorenzano <[hidden email]> wrote:
>>>
>>> hi,
>>>
>>>
>>>> On 13 Jun 2018, at 16:50, Tim Mackinnon <[hidden email]> wrote:
>>>>
>>>> Hi - my second attempt at using Pharo with Git has proven very satisfying (I saw the potential in phase 1, but it was often difficult to understand what was happening and the workflow to use).
>>>>
>>>> One thing that has come up a few times for me however - and its something that using git nicely highlights, there are many non-smalltalk assets in my project that don’t need to live in the image (like Seaside FileLibraries were trying to do) but do need to be versioned and be part of my project. Common examples are server config files, images and even the playground history files that are useful to pull up when on another computer.
>>>>
>>>> It seems that while Iceberg does check out a full project, if I change any of the files outside of the src directory (like edit a .txt file using the crude Pharo file editor), those changes don’t get committed when I do a checkin? Is this on purpose? It makes the workflow a bit trickier to do an atomic commit of a piece of work - and I’m not clear whether this is a conscious thing, or an MVP thing (and it will come later).
>>>
>>> workflow is tricker because you are expecting iceberg to talk with the local working copy and to handle that WC.
>>> what happens in fact is different: iceberg treats the image as a working copy itself (it has its own “stage” area) and what you have in disk is like a separated WC.
>>>
>>> at least, this is the metaphor we are using now, because we cannot realistically handle/control what is in disk since it can be anything.
>>> So, instead having this picture in mind:
>>>
>>> Image -> Disk -> Git blob (database)
>>>
>>> you need to have this other:
>>>
>>> Image \
>>>      Git blob (database)
>>> Disk    /
>>>
>>>
>>> you will see as soon as you change the mental image, your problems are gone ;)
>>>
>>> cheers!
>>> Esteban
>>>
>>> ps: diagram before is not exactly as it is since the image actually writes into disk first, but this is an implementation detail we would like to remove in the future, even.
>>>
>>>>
>>>> As mentioned above, I was also thinking it would be nice if I could checkin some of the play-xxxx/*.sh files to essentially keep some of that history synced between environments (or team members?).
>>>>
>>>> It strikes me that this is the kind of thing that git integration should bring to us?
>>>>
>>>> I can overlay my copy of IntelliJ on top of my local iceberg directory and then use it for checkins - but then I still have the atomic problem, as its only when I commit that tonel files are written out onto the file system for me to checkin along with any other assets I’ve changed. Does anyone else have a good workflow for this? What do you guys do?
>>>>
>>>> Tim
>>>
>>>
>>
>>
>
>

--

Guille Polito Research Engineer Centre de Recherche en Informatique, Signal et Automatique de Lille CRIStAL - UMR 9189 French National Center for Scientific Research - http://www.cnrs.fr Web: http://guillep.github.io Phone: +33 06 52 70 66 13

Am 15.06.2018 um 10:29 schrieb Tim Mackinnon <[hidden email]>:

In many parts of the dev world - every commit should be shippable, meaning atomic - particularly with the use of CI systems (Travis, Gitlab etc) that build on every commit. 

Of course you can choose to work a different way - but our tools shouldn’t force a very common practice to be awkward, should they?

I don’t understand why many of you guys are so anti such a common thing. In fact our community pioneered this way of working - a build machine where you loaded all your code and ran all your tests before committing.  Also, a large number of agile teams will council to work on master (and avoid branches - a technique we also pioneered).

In today’s world where we live in a wider polyglot stack this is one of the reasons why git is so attractive as it should easily embraces that.

I get that it’s not quite possible yet - but there are some decent ways to accommodate this (a staging operation, or perhaps a better way to make use of the #addToIndex: which I will experiment with).

Anyway, I’ll shut up - and possibly will learn how to implement it myself. Thinking more, I may also be able run with commit and no push , which I will try - as it’s probably the easiest mechanism if the CI will recognise that (I think it will).

Thanks for the enlightening conversation, and as always the hard work on iceberg that made this conversation even possible.

Tim

Sent from my iPhone

On 15 Jun 2018, at 07:11, Norbert Hartl <[hidden email]> wrote:

Am 14.06.2018 um 13:12 schrieb Thierry Goubier <[hidden email]>:

Hi Norbert, Tim,

2018-06-14 11:33 GMT+02:00 Norbert Hartl <[hidden email]>:

Am 14.06.2018 um 10:30 schrieb Tim Mackinnon <[hidden email]>:

Hi - yes I’m pleased you check out the entire tree, although currently it’s a bit confusing that you do (fortunately this does give the possibility that we can checkout images and other resources that an Pharo application might rely on - without having to resort to the Seaside FileLibrary trick).

However my concrete case was that I have a gitlab ci pipeline and next to my src directory in my project I have a config directory that has some Nginx config for my teapot app. If I add a teapot route, I also need to adjust that config and check both changes in together. I can’t easily do that now?

I can modify /config/app.nginx either in another app (intellij) or even in the simple Pharo text editor, and the I can add my new route in my DemoApp>>createRoutes method but how do I check them in together so my pipeline will build atomically?

Iceberg hasn’t written out the changes yet, so IntelliJ can’t see them to do a commit, and iceberg ignores the parallel /config directory (that it checked out). So it’s a catch 22.

This is why I suggested maybe we could specify safer (textual) directories that iceberg might also checkin? OR we have a Stage command in iceberg that does everything that commit does up to the point of actually writing to the repo - then I could jump to IntelliJ and do the final commit there and use its tools to manage non Pharo stuff (until we can build more)?

Does this make sense?

I don’t understand why you are so eager to have everything into one commit. Usually the tension is rather have small commits. What is the problem of having two commits for this?

A single commit allow one to make sure that both the smalltalk code and the external resource are well in sync, and that you may not ending up in a situation were at commit j has data format v1 and smalltalk code for format v2, because it is in commit j+1 that you rewrote the data in format v2.

Yes, sure but that is only a problem if you lost control over which commit goes into action. I think the problem is in the deployment then. Putting integrity constraints on commits is IMHO the wrong level deployment granularity

Norbert

I had that issue recently with a Pharo / FPGA project with a Pharo package generating state machines to be integrated in a Verilog FPGA design with C code for the drivers, the softcore in the FPGA, and test programs, and both Pharo and the C/FPGA working out of the same test data... And that whole thing getting regularly out of sync.

IMHO: I'd model a concept of "multi-lingual projet" for that sort of things, where one can list, in Pharo, Monticello packages and external files or directories. Whatever the technology you use (OSProcess, libgit, whatever...) it is easy then to commit all this in a single pass. On a per-package basis, I'd see the package manifest as a suitable place for listing external resources. On a per-project basis, a possible place could be the manifest for the BaselineOf the project.

Thierry

 

Norbert

As an aside - I’d really like to checkin in the play-xxx directories (the .ph files) as there is often useful playground stuff I’d like to access on my home computer. We can’t do that easily at the moment either.

Tim

Sent from my iPhone

On 14 Jun 2018, at 09:12, Guillermo Polito <[hidden email]> wrote:

Just to complement Esteban's answer:

- Iceberg checks out in disk more than the src directory because you **may** want to edit files from the command line, and after long discussions we did not want to forbid that.

Actually, just to put everybody in perspective, at first the idea was to not have a working copy in disk at all, but just hit to the blob.

Imagine is nowadays we are a bit alien, that would have been worst :)

- About checking in files. I'd like to understand what you mean exactly.

  - Do you want to load them into memory?

    This would be the "more consistent" way to do it, following the "the image it its own working copy" metaphore.

    This would allow us to, for example, share an image and transparently share resources with it (without requiring to clone).

    But this would have some impact in memory consumption and add stress to the GC, right?

  - Or do you mean to ask like any other Git client and show you the file differences between the working copy and the git index?

    The problem with this approach is that we will have some treatment for pharo code and some different treatment for non-code...

    If I do a change to a class, the change is kept in the image. But if I do a change to a file, that change is not kept in the image!

Also, as Esteban says, having an IDE with support for files would mean that we would need good tools to edit in-memory files (not only text files, right? but also any kind of binary file...)

So far we cover the bare minimum that allows us to *not lose* changes :)

On Wed, Jun 13, 2018 at 11:07 PM Tim Mackinnon <[hidden email]> wrote:

> yeah… but is a lot of work and no time just right now.
> long term, it would be cool to manage everything from iceberg.
> but reality check, is a huge amount of work so it has to come step by step.

Fair enough - its pretty cool we’ve got this far, and I guess the onus is on the rest of us to learn more about how its done and see if we can contribute more somehow. I really appreciate the love you’ve already put into this - it works far better than I think we even realised it could.

Tim



> On 13 Jun 2018, at 21:55, Esteban Lorenzano <[hidden email]> wrote:
>
>
>
>> On 13 Jun 2018, at 22:44, Tim Mackinnon <[hidden email]> wrote:
>>
>> Esteban - so I don't then understand why iceberg (usefully in my view) checks out more than the src directory, if it’s only focusing on the Pharo blob?
>>
>> I’m guessing that by knowing where the src is, you are just committing that part of the tree with libgit?
>>
>> Perhaps from a pragmatic first step you might consider letting us add a second safe resources directory that you could check in atomically as well (on the understanding all bets are off if it goes wrong?)
>>
>> OR could we have a check in mode does all the add/remove operations and writes to disk but then let’s you drop to the command line/other tool to add any other files and do the final commit?
>>
>> I just feel like you/we are so close to something that works a bit more broadly and embrace the wider world.?
>
> yeah… but is a lot of work and no time just right now.
> long term, it would be cool to manage everything from iceberg.
> but reality check, is a huge amount of work so it has to come step by step.
>
>>
>> Tim
>>
>> Sent from my iPhone
>>
>>> On 13 Jun 2018, at 21:28, Esteban Lorenzano <[hidden email]> wrote:
>>>
>>> hi,
>>>
>>>
>>>> On 13 Jun 2018, at 16:50, Tim Mackinnon <[hidden email]> wrote:
>>>>
>>>> Hi - my second attempt at using Pharo with Git has proven very satisfying (I saw the potential in phase 1, but it was often difficult to understand what was happening and the workflow to use).
>>>>
>>>> One thing that has come up a few times for me however - and its something that using git nicely highlights, there are many non-smalltalk assets in my project that don’t need to live in the image (like Seaside FileLibraries were trying to do) but do need to be versioned and be part of my project. Common examples are server config files, images and even the playground history files that are useful to pull up when on another computer.
>>>>
>>>> It seems that while Iceberg does check out a full project, if I change any of the files outside of the src directory (like edit a .txt file using the crude Pharo file editor), those changes don’t get committed when I do a checkin? Is this on purpose? It makes the workflow a bit trickier to do an atomic commit of a piece of work - and I’m not clear whether this is a conscious thing, or an MVP thing (and it will come later).
>>>
>>> workflow is tricker because you are expecting iceberg to talk with the local working copy and to handle that WC.
>>> what happens in fact is different: iceberg treats the image as a working copy itself (it has its own “stage” area) and what you have in disk is like a separated WC.
>>>
>>> at least, this is the metaphor we are using now, because we cannot realistically handle/control what is in disk since it can be anything.
>>> So, instead having this picture in mind:
>>>
>>> Image -> Disk -> Git blob (database)
>>>
>>> you need to have this other:
>>>
>>> Image \
>>>      Git blob (database)
>>> Disk    /
>>>
>>>
>>> you will see as soon as you change the mental image, your problems are gone ;)
>>>
>>> cheers!
>>> Esteban
>>>
>>> ps: diagram before is not exactly as it is since the image actually writes into disk first, but this is an implementation detail we would like to remove in the future, even.
>>>
>>>>
>>>> As mentioned above, I was also thinking it would be nice if I could checkin some of the play-xxxx/*.sh files to essentially keep some of that history synced between environments (or team members?).
>>>>
>>>> It strikes me that this is the kind of thing that git integration should bring to us?
>>>>
>>>> I can overlay my copy of IntelliJ on top of my local iceberg directory and then use it for checkins - but then I still have the atomic problem, as its only when I commit that tonel files are written out onto the file system for me to checkin along with any other assets I’ve changed. Does anyone else have a good workflow for this? What do you guys do?
>>>>
>>>> Tim
>>>
>>>
>>
>>
>
>

--

Guille Polito Research Engineer Centre de Recherche en Informatique, Signal et Automatique de Lille CRIStAL - UMR 9189 French National Center for Scientific Research - http://www.cnrs.fr Web: http://guillep.github.io Phone: +33 06 52 70 66 13

On Jun 15, 2018, at 2:05 AM, Esteban Lorenzano <[hidden email]> wrote:

On 15 Jun 2018, at 10:29, Tim Mackinnon <[hidden email]> wrote:

In many parts of the dev world - every commit should be shippable, meaning atomic - particularly with the use of CI systems (Travis, Gitlab etc) that build on every commit. 

but then, you mean every *push* should be shippable :)

that’s something I agree :)