openerp-expert-framework team mailing list archive
Mailing list archive
Making our way out from the bloated extra-addons repository
Dear OpenERP experts,
Some of you might not have already think about it, but those who have all
came to the same conclusion I will re-explain here:
The system of the single extra-addons branch should now be avoided because
it doesn't scale in term of code and community management:
1. enabling or restricting commit rights doesn't scale: you may want to
allow commit rights to anybody of your team for you little non critical
extra-addons but the other users may want to be sure the extra financial
addons they are using in production are rock solid and not screwed
overnight by some beginner.
2. extra-addons receive commits for a huge number of addons. Hence it's
really challenging to track the code evolution of the critical extra-addons
3. because there are so many extra-addons, the branch is huge! Often you
just need 2 or 3 extra addons and you are still forced to download hundreds
of Mb of code. This is specially discouraging contributions in all the
countries where Internet isn't so fast...
4. because there has been no quality management of those modules, the
extra-addons branch is full of poor quality non reusable, non migrated
modules. Probably 75% of the extra-addons modules are broken and not
reusable outside from the initial company or even POC they where developed
for. Worse than that, at the beginning, say back in 2008, Tiny and Axelor
were putting almost all their POC modules with not enough quality to
qualify as addons in the extra-addons branch, resulting in garbage modules
lagging the extra-addons quality until today. Following this "example", few
companies refrained themselves from pushing their low quality modules in
the extra-addons during those years.
5. often you need to improve some extra-addons for several reasons,
1. adapt it further to the current stable OpenERP version your are
2. add new features for the company you are working for if you think
it makes sense to have that feature inside the module
3. fix a bug in a non backward compatible way
4. refactor the code to make the module more compatible with some
other modules or localization
In all those cases, the proper way is to create a new "feature branch" and
eventually merge it back into the "stable" branch. But because things
aren't so mature yet, often changes are not backward compatibles and it's
better to stay in a new branch people should explicitly pick, while keeping
the "stable" branch for really stable things people can update without
special care (ideally). If a branch has many modules like the extra-addons,
then it creates many parallel versions of the same modules which is a real
hassle to maintain (addons_path won't do the job etc...)
Some of us know that for some time already, so instead of bloating the
extra-addons with new modules, we have been creating new modules in new
branches since one or two years.
But now, we should go further, we should move the extra-addons we use often
out from the extra-addons into smaller manageable branches.
Because ERP means big money involved, OpenERP contributor copyrights are
often abused while copyrights tend to be an essential pillar of how open
source works. So it's essential that the bzr history of the extracted
module don't get lost and that real authors can be traced back (if that get
lost, it will increase the risk somebody comes and change the license
claiming they are the only author).
At Akretion we developed some semi-manual script to do that and replay the
module commits in the new branch. I'll pass it soon in this list.
Also what's the solution: one module per branch or several modules in a
Well eco-systems that use well working module managements system tend to
use the one module per branch approach as one can easily rebuild at anytime
all the dependency tree with the proper module version that have been
tested to work together.
But given the OpenERP approach of not dealing with module version
will not be easy to spot the compatible modules versions (that information
is not even stored!). Imagine if you used 20 extra-addons, you would now
need to do bzr pull in 20 different branches and ensure manually you are
taking the proper branches and updating to compatible revisions...
So given this situation, at Akretion we think it can still make sense to
group some related modules inside the same branch until possible better
module management appears and possible further branch split.
*Grouping modules together*
Then comes the grouping problem: what modules go together into the same
branch? Well, unfortunately there won't be some clear rule, it has to be a
tradeoff. One could think just put everything that depends on sale
together, everything that depend on stock together. Unfortunately this
isn't that simple: there are modules having multiple core dependencies and
many border line cases (where would the stock_rma module fit: stock or
CRM?, where would the account_fiscal_position_rule module fit: account or
So let's start discussing those groupings if you like. I'm taking several
factors into account:
- functional domain
- avoid to leave just one required module alone in a new branch to
checkout if possible
- usage in localizations...
Sebastien Beau (Akretion) already extracted the fiscal rule modules
together in a new project/branch:
It contains those modules who where extra-addons originally:
We developed those above modules for the Brazilian localization. But in
fact they are potentially useful for managing the fiscal positions in all
countries which are federations (possibly even the USA) and are useful for
international fiscal operations. Recently we started using them more for
ecommerce projects in Europe and Canada for instance to select the right
tax accounts even if the ecommerce front-end (like Magento) already pass
the proper VAT ratio to OpenERP.
We should continue that migration process further and ideally stop
depending on the extra-addons for OpenERP v7 projects.
Also, by removing the dependencies on that fat extra-addons repository, we
also make automated Continuous Integration testing easier. At Akretion we
are experimenting with Github mirrors and Travis-CI triggers to run
automated test at every commit. It's not that we don't like the Runbot, but
we can run the same tests for free on any branch and easily add support for
next testing suites like OERPScenario test suites freely. Also, OpenERP SA
re-affirmed they won't care about module dependencies, but this doesn't
match what we do at Akretion (who here already tried to install the proper
Magentoerpconnect dependencies?), so we will need to be able to specify the
proper branch dependencies for all the testing and have no faith the Runbot
can do that when the version dependencies will be missing from the module
format. Well, instead free open source tools like Travis-CI provision a new
Virtualbox machine at every run and do a fresh repository checkout, so by
depending on lighter repositories, we also make all that easier to happen.
This is part of the road to a better quality of those OpenERP modules.
I would appreciate your view on that question and will post again soon to
submit a new concrete module extraction proposal for the financial
extra-addons used both by the Spanish and Brazilian localization at least.
Founder and consultant
+55 21 2516 2954