Repo and branch: https://github.com/tsuburin/etckeeper/tree/feature/delegate-ignore-check-to-git

Summary

This PR fixes the problem that empty directories in .gitignore are not actually ignored in .etckeeper.

Description

The function filter_ignore in pre-commit.d/30store-metadata doesn't filter empty directories. For example,

[root@ankou ~]# cd /etc
[root@ankou etc]# mkdir empty
[root@ankou etc]# echo empty >> .gitignore
[root@ankou etc]# etckeeper pre-commit
[root@ankou etc]# grep empty .etckeeper
mkdir -p './empty'
maybe chmod 0755 'empty'

There are two points in the current script that cause this problem.

First, filter_ignore is not applied when generating mkdir -p lines at line 71-72,

find $NOVCS -type d -empty -print |
    sort | shellquote | sed -e "s/^/mkdir -p /"

So, no directory is excluded here, regardless of the contents of .gitignore.

Second, filter_ignore doesn't handle directories correctly. Here I found two problems.

One is that git ls-files puts a trailing slash at the end of each directory while find doesn't.

[root@ankou etc]# git ls-files -oi --exclude-standard --directory
empty/
[root@ankou etc]# find . -path ./.git -prune -o -type d -empty -print | grep empty
./empty

So if we use this approach, when comparing them at line 33,

sed 's/^\.\///' | LC_CTYPE=C grep -xFvf "$listfile"

we have to make them in the same format by some means like sed 's-$-/-', as proposed in Do not recreate ignored empty director.

The other problem is that with --directory option git ls-files outputs only the uppermost ignored directories, so nested directories cannot be filtered. Let's say we have an entry /xxx/ in .gitignore, and /etc/xxx/ has a directory yyy in it. In this case, git ls-files -oi --exclude-standard --directory outputs /etc/xxx/, but not /etc/xxx/yyy/. So, later when we apply filter_ignore to the result of find at line 86,

find $NOVCS \( -type f -or -type d \) -print | filter_ignore | sort | maybe_chmod_chown

/etc/xxx/yyy is not excluded. I think this problem is a big burden, so I propose a different approach in my PR.

My PR fixes these points. For the first point, I just added filter_ignored at aforementioned place. For the second point, I rewrote filtering logic using git check-ignore. This command tests each path if it's ignored regardless of whether it's a file or a directory, so we can simply filter all find-ed paths. Additionally, the command outputs the paths without modifying, so we no longer need to modify them;; by sed.

Related posts

• https://etckeeper.branchable.com/todo/metadata_ignore_filters_do_not_work/
• https://etckeeper.branchable.com/todo/Do_not_recreate_ignored_empty_directory/