If you get a chance to read these articles I highly recommend them
http://www.alwaysagileconsulting.com/organisation-pattern-trunk-based-development/
http://endoflineblog.com/gitflow-considered-harmful
http://endoflineblog.com/follow-up-to-gitflow-considered-harmful
Jacques
Le 12/05/2015 19:28, Adam Heath a écrit :
Nice. This is quite thorough. There is an option missing. SVN committers who use git offline. In this case, their changes can be published as
primary SVN branches, for collaboration.. See OFBIZ-6271 in JIRA, and as an SVN branch, for an example.
I've read through most of what follows, and am in agreement, but I'm dealing
with hardware problems, so I need to let it sink in first.
On 05/12/2015 04:43 AM, Taher Alkhateeb wrote:
Hi everyone,
This email refers to the thread for voting to move to git (link at bottom) in which the vote decision was to delay and elaborate on the workflow
first. I am not well versed in ASF guidelines and appreciate any help and feedback and also please note some of the below is my opinion and not
necessarily 100% factual.
## First, identified problems
1. patches can quickly be outdated if not applied quickly
2. big patches are hard to audit and not desired nor preferred and It is hard to break big patches to smaller ones because if any of those patches
is outdated or modified then everything needs to be re-patched
3. to collaborate with other people (non-committers) freely on big features, we need a separate branch. On svn this is lengthy and heavily
controlled. If we create a git repository then we need to constantly update from svn and merge . Another solution is to clone the ofbiz read-only
git repository but then there are some patch issues to convert them to clean svn patches (I faced a few including things like white space)
4. a lot of _local_ offline freedom to branch, merge, commit, share and experiment cannot be easily done without initiating a local git repository
which triggers the other problems identified above.
6. There are too many public branches in the repositoy. Some are not active nor
complete and quite old
## Second, how does git provide solutions
So, adopting git in relation to the above mentioned problems solves them as
follows:
1. even if a patch gets outdated, I can easily recreate it by switching to a branch that I created and has the work (e.g. OFBIZ-12345), merging
everything from trunk and re-patching
2. to allow for proper feedback by community, a pull request can replace a big patch and that request can hold an X amount of commits each with its
own message and diff details. If changes happen to any of the commits, then reconciling that into the code base is minor, you just branch again, do
it, and merge. Furthermore, I suggest to follow the guidelines which recommend rebasing before pushing to a shared repository to keep a nice linear
history as much as possible as shown here -> https://git-wip-us.apache.org/docs/committer-practices.html
3. large features can be done in a remote repository in github or bitbucket
with pull requests when complete and ready for review.
4. the issue is immediately solved with git which is not only local but much,
much faster
6. We do not need to pollute the main repository with branches if we decide on a distributed model like git with remote repositories to contribute
to the project with pull requests.
## Third, proposed workflow
I will make a distinction between small features / bug fixes and large features.
### small features
Small features follow the exact same workflow that currently exists in svn. You do your work, diff it, and attach the patch to a JIRA and request a
commit from one of the committers.
### large features
For large features usually multiple people need to collaborate on a separate
branch. Here is where git shines and the distributed model kicks in:
1. A JIRA is created for a large feature
2. The team (not necessarily having a committer) creates a remote repository which itself may have many branches with the master branch having all
the work agreed upon and merged (actually, rebased)
3. The collaboration for this branch happens in the JIRA including discussions,
comments, and even links to the commits etc ...
4. A request is made to a committer to make a pull request from the repository after reaching a certain milestone with consensus from the community
of course
5. Here, for extra safety, the branch model may have a trunk and a develop branches. Everything is pulled to the develop branch and trickles down
to the master branch after thorough and proper testing.
The above workflow can also adhere to the now famous Vincent Driessen git branching model found here ->
http://nvie.com/posts/a-successful-git-branching-model/
I am not sure whether this proposal is enough or correct so I appreciate your
guidance and feedback to fix whatever needs fixing.
Taher Alkhateeb
original voting thread:
http://ofbiz.markmail.org/search/?q=move%20to%20git#query:move%20to%20git+page:1+mid:p62ofojcbb3oyoi4+state:results
