It's a standard to use the imperative mood in writing commit messages - the Linux kernel enforces this[0], as do a lot of other projects.

It makes more sense if you consider a commit to be an independent "thing" in a list, and you follow the list from start to finish to end up with the current repository state, each commit being a command.

[0] http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.g...

Because generally you're reading commits that happened before and you want to know what happens when you apply the commit.

It goes something like this.

"Hmm, I wonder what would happen if I merge 1231knf?"

It would "Add email integration on /email route"

Otherwise it would be something like

"What happens when I merge this commit?"

"Added email integration on /email route"

It's considered as best practice when using Git to write in present tense.

That's one interpretation. Another is you're looking at git log and it's all historical. The best practice bit seems to boil down to "because Linus said so." The rest of the git commit best practices seem to fall into a similar boat.

I'm of the mindset a team can be mature enough to pick what works best for them. If the idea of a commit message is to convey useful info, imposing hard rules like 72 chars or tense seems a bit draconian. The worst change GitHub has made IMHO is truncating messages (particularly because the truncated message often ends up longer).

The reason for the 72 character limit is so that shortlog messages on an 80 character terminal which are indented by one tab aren't truncated, as I understand it.

I get that. And in certain scenarios where even pagers aren't available require that. But I'd expect that team to handle it, not for it to be a general purpose rule. In contrast, a Nokia postage stamp feature phone can display a message more than twice as long (SMS - 160 chars). Limiting to 72 chars across the board seems to be more dogmatic than anything else from my POV.

I think that the boat already sailed, there are many tools that assume that shortlogs are 72, and can be displayed on a single line, and format accordingly, including GitHub for instance. Besides, whether it's 72 or 100, it makes sense to force people to have a concise summary.

GitHub's approach is one reason I've started using BitBucket more. GitHub didn't always used to do this. I've assumed they started so they could have fixed height commit displays in their commit list. It seems weird to force this on people otherwise.

I believe the motivation is explained here - http://tbaggery.com/2008/04/19/a-note-about-git-commit-messa...

Write your commit message in the imperative: "Fix bug" and not "Fixed bug" or "Fixes bug." This convention matches up with commit messages generated by commands like git merge and git revert.

That mostly just passes the buck to why do merge and revert use the imperative? Simply being consistent with them doesn't seem like a very strong argument.

I tend to use that style because it matches git's generated commits ("Merge ...") and it saves a few characters. I found that when using a different tense ("Adding" or "Added") I would run out of characters more often.

I thought it was very meta amusing that the author was trying to tell users how to write commit messages when the number one rule of commit messages in my book is explaining WHY the commit exists. Why does this patch change this or do that. So it was amusing when he lists commit rules without explaining why they are rules. But I was saddened when I got to the end and missing from the list is 'explain why'.

Why did you 'Add feature...'? Why did you 'Move cursor to...'?

I did stuff // Increment i by 1

Why I did stuff // Increment i by 1 to not have a off by 1 bug

I see three different types of commits in my codebase, usually.

* The big fix: This is the commit which usually is too complex to be explained in a 72-characters message. Usually it refers to an issue number on the project's issue tracker, which in my case is Github.

* The medium fix: This is a commit whose "why" fits properly into 72 characters, i.e. "Add classes to validate user input only on Friday".

* The small/obvious fix: This is a very small commit, usually fixes obvious issues like CI builds, misspellings in doc files, missing semi-colons, etc etc.

In the last case, I find it appropriate to write the "how" rather than the "why" in the commit message, since the why is already obvious. i.e. "Add missing comma on object identifier".

The "why" can live in the bug tracker, as a feature or fix may have several commits attributed to it.

For larger centralized projects, I have found a requirement on issue numbers in commit messages to work well. It makes a whole lot of management issues easier.

If issue numbers are the only requirement, is it okay if every commit only includes an issue number and no text at all? If not, why not? If they say something should it be the exact same thing as the issue title?

IMHO, no. You should be able to read a commit log and understand what has been done. But if the issue titles are good enough they could absolutely be used as commit messages. Many issues leads to multiple commits however, so what you want for message is a description of the individual commit.

>the number one rule of commit messages in my book is explaining WHY the commit exists

The commit message should say what changed, not why. Comments are for explaining why the code does what it does.

Good commit log: http://git.savannah.gnu.org/cgit/guix.git/commit/?id=fb519bd...

The diff shows you what changed. The commit message is a comment on the diff and absolutely should say what it's for.

This enables you to easily compile release notes, for example.

Yes, it should say what it's for, but it shouldn't need to explain itself any further. Sorry if I didn't make that clear. For example, if someone refactored something, the commit log can simply say "Factorize foo", it need not explain why they were refactoring in the first place.

Comments tell why the code does what it does. Commit messages tell why the code does something different than before.

Example comment: i++; // because server starts counting at 0 (not 1)

Example commit: Our customer switched to bar (from foo), which starts counting at 0 instead

That works as long as you're adding code. But what about when you're deleting code? I absolutely want to know why you've removed something.

Angular has similar rules.

Maybe it's because "Add" is shorter than "Adding" or "Added"? When I first worked on a project that followed the Angular pattern, the guy ramping me up said "Just imagine that you're commanding the computer to do something, and use that tense."

The past sounds good, until recently I was writing my commit messages using it, but when you start making pull requests on GitHub, you realize that it makes a lot more sense in general to use present for the titles of your commits.

A commit can be seen as a patch, which has a single purpose: "Add...".

"/email route: Email integration added" is my favorite way. Same thing with code naming, I always put the widest context first, then narrow down. With this habit I spend a lot less time thinking about names.

Say you want to apply a certain commit to your branch, it makes more sense to "add" a feature than to "added" a feature.

What does a commit do? It adds/fixes/breaks, it doesn't added/fixed/broke.

Thats one way of looking at it. Another way is - What did I(the developer) do in the commit? I added/fixed/broke, right?