Question

I am using git, trying to sync to the first commits of each month for the last year. I want to be able to go back month by month, on the first day of the month, ON THE FIRST SECOND OF THE MONTH. or at least from some constant “seconds” value. So far I have this:

$(git rev-list --before "$(date -d "$(date +%Y-%m-01) -$i months" +%Y-%m)-01" -n 01 HEAD)

This obviously does not include a constant seconds value. As it stands, running this script and then running it again an hour later returns two different sha1’s because it is going back x months FROM THE EXACT TIME at which I run the script. I want the returned sha1’s to be the same no matter when I run this script. Does that make sense? Any ideas?

Answer 1

First of all, you have a typo in your command. It has to be --before="$(..., you are missing the =.

The behaviour of date is not the problem here. date +%Y-%m-01 will return something like 2011-10-01 as a string, there’s no additional time information included. So the second date call will decrease it by the $i number of months and will also return a sting of the format like 2011-09-01. Besides that string no additional information is passed to git-rev-list as value of argument --before=.

There are some things you have to consider using git-rev-list:

1. When using HEAD you’re always referring to the current branch. So when you for example checkout one of the obtained commit-IDs your HEAD will change. You maybe want to use master or any other branch name as reference instead.
2. Git has no temporal order of commits. You can have a commit authored on 1st of August hierarchically after a commit dated 1st of September. This will result in confusing outputs when using --before or --after arguments since they rely on the timestamp of the committer field.
3. The timestamp of the committer field may also be misleading. When your branch doesn’t have a linear history, it’s hard to tell whether a commit has been part of the branch in a repository at a certain point in history. The author may have pushed/merged his branch X months after doing a commit, so it have not been visible to others.

Considering all the said, the following command works for me:

$(git rev-list --after="$(date -d "$(date +%Y-%m-01) -$i months" +%Y-%m)-01 00:00:00" master | tail -n 1)

This will return the ID of the first commit after the first of the given month. (This commit was not necessarily made during that month, maybe there were no commits in that month anyway.)