Contributors mailing list archives
Re: Procedure to create 16.0 branchesby
Acsone SA/NV, Stéphane Bidoul
Let me answer what I perceive has the most important points:
> Also derived from the first point, it's the lack of homogeneity in the processes. With the current one, everything follows the same procedure. This is VERY IMPORTANT: to have only one way of doing everything, not having to know different processes depending on the state of the branch.
Please note that the current procedure with "git format-patch | git am" will continue to work as before: it will just pick up missing commits. There is no obligation to learn new tools.
> Modules being migrated with a jump of version will have the same problem of not having any of the commit history.
That is true but this will fade away over time.
> Rebel modules configuration and similar CI tricks may cause problems on new branches.
All the CI is now configured with copier questions, and we will reset the rebel module groups when creating new branches. So that should not be a problem.
On Wed, Jul 20, 2022 at 4:52 PM Pedro M. Baeza (Tecnativa) <email@example.com> wrote:
Returning back to this approach is a mistake IMO, as it has more cons than pros. First of all, one of the advantages that you mention is not correct, or at least, depending from the perspective, that is the "slow git repo growth": if you download a single new branch, it will cost a lot more, as it has all the commit history from all the modules. This gets even worse when the modules are deprecated (more on this later).
And about the drawbacks, there are still bad interactions with modules with previous versions code due to Python or ORM peculiarities (check for example this PR for fixing a code that shouldn't be executed not being the module installed), and for sure more will appear. Going to the process of discovering these problems and fixing this is an unneeded maintenance burden. The discoverability will have the same problem, as people don't get used to having useless folders an d to need to go to that file to check it or open the manifest. Even more, I think the OCA apps store module will have problems with this approach (but that's only suspicions).Now let me summarize again the rest of the cons that were the reason to change to this approach:
- When the branches are created for the new 16.0 version, activity in the prior branch 15.0 may happen until the migration of the module is done, but the 16.0 branch won't reflect these new commits. There's a high risk of the contributors of the migration not including latest changes, or to work on the deprecated source code and come to OCA with their pull request to be discouraged by one qualified reviewer saying that they need to rework the migration including the missing commits. Forcing and teaching people to use tools like oca-port is a high entry barrier. Someone has proposed to develop a bot for automatically doing these forward-ports, but it will have lots of problems as well:
- First of all, it needs to be developed, which is not an easy task and needs to be done. Adding extra tools for something that is not needed with another approach is not the convenient way to go.
- If there's already a migration PR proposed, the new commits won't be there, and it will require a rebase by the migrator (again more contribution friction).
- Those extra commits will lose the security signing that may be present in previous branches. The only way to not lose them is that the same author makes the commit signing it in the new branch.
- There can potentially be the same CLA problems.
- When the bot does the forward-ports, CIs may be red due to external reasons, but then these commits won't never reach the new branch unless manual actions are done.
- This will increase the size of the repository the same, as no common ancestor and then different SHA.
- Derived from this, it requires that the reviewers alw ays need to check if the migration pull request contains the latest commits. With the previous approach, there are less chances that they are not included, and we light up the reviewing process.
- Also derived from the first point, it's the lack of homogeneity in the processes. With the current one, everything follows the same procedure. This is VERY IMPORTANT: to have only one way of doing everything, not having to know different processes depending on the state of the branch.
- It will generate extra burden for deprecated modules, requiring to make a pull request to remove them, and creating a very big diff commit of such removal, increasing the branch and repository size (remember that there will be at least one commit adding the module, and another removing it, so double diff).
- Modules being migrated with a jump of version will have the same problem of not having any of the commit history.
- Rebel modules configuration and similar CI tricks may cause problems on new branches.
- In the past, having uninstallable modules has cost a lot of problems. Remember the static folder problem that loads the Python files, the switch from version 2 to 3. Now they are not present, but nothing prevents a new side effect from happening in the future. Better to have the house cleaned.The fact that some of you are using this approach in some repositories is not representative, as that repositories have a very limited scope (few modules), and very controlled by few contributors, but try it in OCA/sale-workflow or OCA/purchase-workflow...Please reconsider your approach. If not, we at Tecnativa don't know if we can afford this process in our contributions to OCA.Regards.