Hi Stian, inline...

On 2016-03-11 06:53, Stian Thorgersen wrote:


On 10 March 2016 at 22:17, Peter Palaga <ppalaga@redhat.com> wrote:
Hi Stian,

just a few minor points. Please note that I am speaking primarily from the standpoint of somebody who often looks into the history of KC branches, because Hawkular (where I am primarily active) uses KC. I am not a regular contributor to KC.

(1)
> With JIRA key in commit and PRs linked in JIRA we get a good history
> that is useful in the future to identify what was changed and why.

 To see *what* was changed, I do not need to go to Jira. Git history is the ultimate source of that info.

As for the *why*, the project has to coin some convention. From what you say it seems that it should be Jira, right? Yes, that is one of the legitimate choices, but not the only possible one. Git commit messages may serve the purpose too and I personally find them much more practical because they are much closer to where I usually start looking. Jira is usually one slooowly loading link further. Therefore, could you please explain why you find Jira better?

One isn't better than the other. They serve different purposes. Git commit messages will tell you what code was changed, while JIRA descriptions will tell you why it was changed. Both are equally important. Making sure you can go from a JIRA issue to Git commits and the other way around are both important. Imagine two examples:

* A user searches for a bug in JIRA to see if the problem he's having has been resolved. He finds the issue and wants to take a look at what changes was made to understand it better.
Yes, I fully agree that the linking from Jira to GitHub makes sense.
* A developer finds some code he wrote 12 months ago was changed, looks at Git history to see who did it, but has no idea why. He then sees the JIRA key in the commit message and opens it in JIRA. There he finds a lengthy description of the bug, including several comments.
No, the explanation why the given commit was necessary may well be placed in the commit message itself. In many projects, this is the default place for that kind of info. Now, given that the motivation for the change can be described in the commit message, the overhead of creating a Jira needs to be justified. Writing the "why" to the commit message is much faster and more developer friendly than forcing contributors to create a Jira for every trivial improvement. I can say for myself that having to create a JIra before fixing some stupid bug substantially lowers my readiness to fix it at all.

The work has to be scheduled the same as all other work.
Yes, planning is fully legitimate within your team paid for the work but again, it sounds rather discouraging to me, an external contributor, when I just want to fix a malfunction in KC that blocks my primary task.


(2) I want tags in branches.
Your tags are outside branches (such as 1.9.x) and that makes it harder to clearly and quickly see what happened between e.g. 1.9.0 and 1.9.1. You are the second project doing it like this out of ~150 Java projects I have ever cloned.

Please consider tagging the usual way.

You've certainly cloned more repositories than I have.. I'm not sure what you mean. 1.9.x tags has to come from 1.9.x branch as master is now 2.x. Can you elaborate?
Sure. This is the "normal" way, where tags 0.17.Final and 0.16.Final are *inside* a branch (master branch in this case):


To get the tags from upstream, I do not need to use --tags with git fetch, because the in-branch tags are fetched also without --tags. Furthermore I can clearly see what happened between the yellow labels 0.17.Final and 0.16.Final.

OTOH in KC, when I forget to fetch with --tags, switching to 1.9.x branch I get only this:


No tags there. Only some hints like "Next is 1.9.2" where the tags are perhaps around. Nevermind, I know I have to fetch with --tags:


Still no tags there, because they are out of the 1.9.x branch that I have checked out. To see them, I have to "Show all branches and tags" in egit. Sadly, that brings also master and other branches, that I do not want to see, but let's try to ignore them, and have a look at what happened between 1.9.0.Final and 1.9.1.Final:


While it is relatively easy to find out how 1.9.1.Final tag relates to 1.9.x branch, I am completely lost on the other end: 1.9.0.Final is a long fork of 1.9.x that I cannot track down visually at all. If 1.9.0.Final and 1.9.1.Final tags were a part of 1.9.x history, my eyes would be much happier. That's it.

Thanks,

-- Peter
 

(3) While I agree that the linking between git history and Jira is useful, saying that every commit needs a Jira does not sound reasonable. There certainly are commits that do not deserve a Jira, such as:
* upgrades of dependencies
* refactorings
* various formatting and license header fixes
I hope most of you will agree here.

I disagree, I want a complete record of what's been done in JIRA. Upgrading of dependencies should not just be done randomly. It has a reasoning behind it (for example a security fix, or to re-allign with WF dependencies, or whatever) without a JIRA you don't have that. Same goes for refactoring, they either should be done as part of an existing JIRA or there should be a JIRA describing why it was done. The work has to be scheduled the same as all other work.

For minor documentation updates (grammar, spelling fixes, etc.) or minor pure formatting sure they don't need a JIRA.

All rules come with exceptions, that's just the way of life. However, by defacto all commits should have an associated JIRA. That's even more important for external contributions.
 

(4) Pro tip for Chrome users:
I created a content script that makes Jiras and BZs on GitHub clickable links. This is what it looks like: http://snag.gy/rq6Ja.jpg How to install: https://goo.gl/bYUbHS

Thanks,

Peter


On 2016-03-10 19:12, John Dennis wrote:
On 03/10/2016 01:03 PM, Bill Burke wrote:
Is there a way to automatically squash everything to your latest
commit?

Maybe not automatically, but the conventional way to squash commits is
with interactive rebase, e.g.

% git rebase -i commit

where commit is the commit id of the *prior* commit you want to start
the squashing from. Git will fire up your defined editor and you just
replace the pick keyword with the squash keyword for each commit to be
squashed. Many examples on the web. It's not automatic but usually less
than a minute from start to finish.

HTH,