gitlint is showing all files as changed in github CI

Gitlint is incorrectly listing all files as changed in CI, not only those that have changed: https://github.com/jorisroovers/gitlint/actions/runs/3014300199/jobs/4935936697#step:14:104

Is this perhaps due to a misuse of refspec? Should we be passing a -1 argument like we do with log?

[jorisroovers]

I realize I didn’t mention earlier that there’s really 2 parts to this.

Initial Commits and the git diff-tree --root flag

An important detail worth repeating is that git itself doesn’t consider files of the initial commit as “changed files” when you do git diff-tree, i.e. for the initial commit, git won’t show you the added files unless you specifically tell it to with the --root flag.

I had added the --root flag to the git diff-tree command that gitlint calls to deal with this, but hadn’t consider the shallow clone scenario.

Since I’ve now discovered that it’s not possible for gitlint to (easily + accurately) determine whether a commit is root commit, I believe gitlint should align with git’s behavior and not show files for initial/root commits unless you explicitly tell it to, with a config option like:

[general]
# consider files of root/initial commits as changed files
# (option name subject to change) 
diff-tree-root=False # disabled by default

You don’t actually have to use a .gitlint file for this, rather you can use the -c commandline flag to specify the same: gitlint -c general.diff-tree-root=True. Gitlint will still read the .gitlint file, but apply this config on top.

By using this construct, you can have an if/else statement in CI that deals with the initial commit separately if you really want it to. I think most users won’t actually care for the initial commit. We could documenting and/or provide templates on how to do this in various CI systems.

Fetch depth

However, this diff-tree-root option still doesn’t correctly solve the fetch-depth=1 challenge in CI, it just provides and all-or-nothing alternative to it. Setting diff-tree-root=True will end up showing all files, setting diff-tree-root=False will show none if fetch-depth is only 1.

The only way to resolve this AFAICT, is to increase the fetch-depth, which is CI specific. I don’t want to include CI specific customizations in gitlint itself, but perhaps we can also make this better using documentation and/or templates.

To reproduce this locally:

# Fetch depth of 1
$ git clone --depth 1 https://github.com/jorisroovers/gitlint.git /tmp/gitlint
$ cd /tmp/gitlint
$ git diff-tree --no-commit-id --numstat -r 01f2fa0
# No output :(
$ git diff-tree --no-commit-id --numstat -r --root 01f2fa0
# All files are listed :(
$ rm -rf /tmp/gitlint

# Fetch depth to 2(or more): fixes this issue
# Whether you use --root or not, only matters for the initial commit (not shown here)
$ git clone --depth 2 https://github.com/jorisroovers/gitlint.git /tmp/gitlint
$ cd /tmp/gitlint
$ git diff-tree --no-commit-id --numstat -r 01f2fa0
1	1	gitlint-core/setup.py
$ git diff-tree --no-commit-id --numstat -r --root 01f2fa0
1	1	gitlint-core/setup.py
$ rm -rf /tmp/gitlint

[chbndrhnns]

So there are some bits that won’t work w/o certain pre-setup.

I get that now thanks to your explanation, so thanks!