Steve Huston escribió:
Hi Manuel,
Looking at the graph of the repository, it seems to me that you're
upgrading the windows branch with patches fetched from the qpid svn
repository. As seen in:
http://github.com/shuston/qpid-port/network
I'm not sure the graph reflects that. What I did was:
- Create the git repo
- Update from Apache svn
- Branch windows
- Apply the patches and additions I put in Jira to the windows branch
Don't trust me, as I'm a git newbie. But looking at the graph, it seems
to me that the windows branch has a set of commits, marked as authored
by tross, gsim, nsantos, etc, related with QPID jira issues as:
QPID-1123: Added a timeout (threading issue is still to be fixed), by
arnaudsimon. I guess that change comes from the qpid repository, but we
have an older master branch. Looking at the graph, what I understand
(can be wrong) is that those changes are applied only to the windows
branch, but shouldn't they appear as applied to the master one?
In the web interface, the master last log is:
QPID-1120: don't reset batch size as part of the messages are not accepted
<http://github.com/shuston/qpid-port/commit/18d7adbf5545fd941a1c246da5d577999109857f>
and then still available. <http://github.com/shuston/qpid-port/commit/18d7adbf5545fd941a1c246da5d577999109857f>
by arnaudsimon, 5 days ago
But in the windows branch, we have, previous to your changes some
modifications I thought imported from the svn, the last one being:
Modified to work with boost-1.32
So, am I understanding it wrong, or shouldn't those changes be in the
master and you should have branched from that?
I thought that the workflow should be:
-Somebody is updating the master, importing changes from the qpid
original repository. The more frequently, the better.
-People working in windows and/or solaris branches, should merge changes
from the master.
So, for example, since the change "Modified to work with boost-1.32",
commit 39c28e7c8b7c0da261eb41494db05c50b0598890
<http://github.com/shuston/qpid-port/commit/39c28e7c8b7c0da261eb41494db05c50b0598890>
is applied only to the windows branch, it's going to be hard to create a
patch against the current svn state. Also, merging from other branches
is more difficult, as I would need to take changes from the windows
branch, and changes could get mixed up.
Again, I'm just guessing. No sure about how the whole think works.
Hmm, I'm afraid I've just repeated the same I've said in the previous
email. So, summarizing, how come that 'Modified with boost-1.32' is
commited into the windows branch and not into master?
About the workflow, I'm not sure about the git commands to achieve it. I
think I've read about an easy way to apply changes from another branch
in the git manual, but don't remember exactly. Will try once the master
advances from the point I branched it.
Best regards.
--
Manuel.
I thought that perhaps it would be better to keep the main branch
synched with the svn repository,and then, merge changes from the
main
branch into the port branches.
I believe that's what I did.
Otherwise, both the solaris and windows
branches will have to read and apply changes from the
external qpid svn
repo. Don't you think it would be better to merge from an
updated main
branch in sync? I think that should be also better to
generate patches
against the qpid repo, as easy as comparing with the main branch.
What do you think?
I agree completely, and that is what I thought I did. If you can see
something I did incorrectly, I'd like to know how to correct it.
Thanks,
-Steve
