Question

In an attempt to gather some statistics about a Git repository, I’m looking for a way to do the following:

• For each commit, execute a command (ex; du -h).
• That command should run from the repository base directory “as it looked like” after the commit.
• The command would ideally have access to the commit hash and time stamp.

One application, expressed in quasi-Bash, would be to run

echo $HASH $TIME `du -hs --exclude=".git" . | awk '{ print $1; }'` >> ../sizeovertime

on all commits to get an idea of the growth of the repository.

(Somehow, it feels like it should be possible to use git filter-branch --tree-filter for this but that looks like a terrible hack to me.)

Answer 1

I don’t see how you could do this without checking out each commit, so that is going to take a while on a large repository.

Here’s how you could go about it with bash:

#! /bin/bash

while read co dt ; do
    git checkout $co > /dev/null 2>&1
    size=$(du -hs --exclude=.git|cut -f1)
    echo $co $size $dt
done < <(git rev-list --pretty=format:"%H %ci" --all --date-order |grep -v "^commit")

Warning: this will leave you in detached head state, on the oldest commit, which is not a nice place to be.