Norbert Hartl wrote on 8. 8. 2018 10:27:
>> Am 07.08.2018 um 16:00 schrieb Guillermo Polito <[hidden email]>:
>> I'll write down some of the reasons of the project's design, like that I can afterwards copy paste it in the wiki :).
>> First, this design did not came up from an egg. We worked on it for about two months. And it is thought to be backwards compatible and manage lots of metacello particularities. It may have things that are perfectible, sure, so let's discuss it.
>> One of the main problems we saw in metacello, and that Iceberg inherited, was the "source subdirectory" thing. This source directory had to be specified in the CLIENT, meaning that every time we clone a repository we should know by heart the directory chose by its developer. Moreover, we lack a standard way to do it, so everybody does as he feels (root directory, src, source, repository, mc....).
>> This has some bad consequences:
>> - once a repository is referenced by some other project, it is more complicated to change its source directory. Imagine that tomorrow we set as standard that all git repos should have the code in src. Then voyage should change. And all its clients too.
>> - Making a typo in the code subdirectory means sometimes super ugly errors from metacello that are difficult to debug and understand (e.g., "Cannot resolve BaselineOfMetacello WTF")
>> Moreover, there was another problem that people started stumbling on: the fact that iceberg got confused sometimes thinking that an empty project was in filetree (to keep backwards compatibility with projects without a .properties).
>> So we decided that for this release we wanted to revert a bit that situation. Think object: let's put the meta-data used to interpret a project's structure inside the project itself.
>> The idea is that:
>> - each project should contain both a .project and a .properties file. The first can contain arbitrary project meta-data (such as the source directory). The second contains the cypress properties, which are needed to correctly interpret the code inside the source directory.
>> - a project without a .project file is an old project and cannot be interpreted, because we don't know the source directory
>> - a project without a .properties file is an old project and is by default transformed in a project with a #filetree properties file
>> - an old project cloned from iceberg detects the missing .project file and gives the user the opportunity to declare it (and then commit it explicitly)
>> - an old project cloned and loaded from a Metacello expression defining a source directory will honnor the source directory defined in the Metacello expression (for backwards compatibility, and we have ~500 tests about this).
>> # About defaults values / forcing the user to define a project
>> First, notice that even when the repositories you load are just marked as "dirty".
>> This is because in memory we add a project to your repository.
>> But you're not forced to commit it.
>> Actually, you can still load packages and baselines from that repository without committing.
>> This is in line with Iceberg's "explicitness". We try to not do any destructive operation without asking the user first (that's why we have several preview windows for pushing, pulling, checkout, merge..., and why contrastingly with monticello we show the committed changes on the commit window...). So, instead of transparently "adding the file" we have decided to modify the project in memory and let the user the responsibility to commit that file.
>> If there's a drawback, is that the repository is marked as dirty. Which is a bit noisy, yes, but still I think it's not so bad compared with the previous drawbacks.
>> To solve this, we could have some default values, yes, and only mark it as dirty if the project does not follow the default value.
>> This could work, but right now all projects use different names for their source directories.
>> So the question is, what would be a good default? I'd like to use 'src' since this is short, well known and less alien (all these in the sense that we do not lose anything and we have a lot to gain by using it).
>> However, not much repositories use 'src' so it will still produce a lot of "noise"...
>> But still! Committing that file is a one-time operation. Once people fix their repositories adding the project meta-data, you will not see them dirty anymore. So we can see this as a transition noise too...
>> Of course, new ideas are welcome. I'll let Pablo and Esteban add their points of view on this too.
> I think I can see what is the rationale behind it but I’m not sure this can be the way to go:
> - I don’t think there can be a „standard way“ of defining source directory. And I don’t think that a tool should enforce this however. I keep frontend and backend code in some repositories together so the source is in my case in backend/source. What does it mean for users not using the „standard“ name?
> - I don’t see why there needs to be a 1:1 relationship between a repository and working copy in pharo. It is like this at the moment already but the .project file manifests this. So it should not be supported to have more source dirs in one git repo? It might be not a good idea that the client has to write the source dir but it opens the possibility that there can be more than one.
Sacrificing very old project (ones not having .properties anywhere)
there's a way where you search in the working tree to find all dirs with
.properties and treating them as (parallel) source roots.
> - My mode of working is to have an eye on dirty repositories because that shows what the impact of your work is. If I have a lot of dirty repositories in my repository list it does not feel good and I don’t want that. Especially for projects I don’t have write access to. How can I change this? I’m not sure that assuming everyone will add these files is a likely one.
In Amber I had similar problem with local.amd.json metadata file. As it
wasn't feasible to think other projects like jquery add one just for
Amber to be happy, I allowed to specify "foreign" metadata as well - so
I can add jquery.amd.json to "inject" metadata to foreign project if it
doesn't have its own.
So, yes, there can be .project.Foreign-Name or something in the root of
your project. As a "transitional" solution until Foreign-Name
maintainers add their own .project, which then takes precedence. And
Foreign-Name's packages / repos won't be dirty (only yours, and only once).
Damien Pollet wrote on 8. 8. 2018 13:53:
> First of all, quick stupid question: I'm currently loading my code with
> gitlocal://./src as the repository URL (my workflow starts in a terminal
> rather than in a Pharo image)
> Should I just remove the /src part, now that my repo has the project
> Also, are more features planned for the .project file? E.g. what about
> storing a default selection for Calypso and the Test Runner in there?
> On Wed, 8 Aug 2018 at 10:28, Norbert Hartl <[hidden email]
> <mailto:[hidden email]>> wrote:
> - I don’t think there can be a „standard way“ of defining source
> directory. And I don’t think that a tool should enforce this
> however. I keep frontend and backend code in some repositories
> together so the source is in my case in backend/source. What does it
> mean for users not using the „standard“ name?
> Sure there can. Look at any ruby or maven project, they all have strong
> conventions for organizing projects and standard config files for
> deviating from those conventions.
> I would have preferred if Iceberg picked one convention (arbitrarily) in
> the absence of a .project file instead of forcing its explicit presence.
> IMHO the choice of default directory per se (be it ./, ./src, ./source
> or whatever) matters less than the fact that there is a convention in place.
> - I don’t see why there needs to be a 1:1 relationship between a
> repository and working copy in pharo. It is like this at the moment
> already but the .project file manifests this. So it should not be
> supported to have more source dirs in one git repo? It might be not
> a good idea that the client has to write the source dir but it opens
> the possibility that there can be more than one.
> I see your point here, but by using separate source directories you're
> sort of creating a hydra project… What I mean here is that the source
> directories are separate, but their histories are tangled. If you want
> separate source dirs it kinda means that you want separate change
> histories, doesn't it? What if the same class has diverging definitions
> in separate directories (I wonder what maven does in that case…)?
There are monorepos. Some people / companies love them.
> On the other hand, separate source directories would be helpful to work
> with git-subrepo and similar tools…
> Damien Pollet
> type less, do more [ | ] http://people.untyped.org/damien.pollet
|Free forum by Nabble||Edit this page|