Use merge-base (three dot diff) instead of direct diff for git-diff-ase (#1653)

[bdsl]

This PR:

• Covered by tests
• Doc PR: Document use of three dot diff site#233

Fixes #1653

Currently when using --git-diff-base=master infection generates mutants for any lines that appear in a direct diff between HEAD and master. This PR changes that to use diff between HEAD and the merge-base of HEAD and master - i.e. the diff you see on a github pull request.

We use the three dot diff syntax for git to get the comparison between the merge base and HEAD. This diff shows work done on the HEAD branch, and should ignore any changes made in the parallel on the base branch. This seems more useful for evaluating the head branch, and fits the description:

infection/src/Command/RunCommand.php

Line 253 in d96aa59

'Mutates only added and modified <comment>lines</comment> in files.',

[maks-rafalko]

hello

1. please, run make cs to fix code style issues
2. seems like there is now an issue with this job: Annotations / Mutation Testing Code Review Annotations 8.0 (pull_request) https://github.com/infection/infection/runs/4932358491?check_suite_focus=true#step:7:9 It says [OK] No files in diff found, skipping mutation analysis. while there is obviously a PHP file in the diff that should be mutated

[bdsl]

Thanks Maks. Would you mind taking a look? I added some temporary debugging code, which shows that the generated command line is git diff 'origin/master...HEAD' --diff-filter='AM' --name-only | grep src/ | paste -s -d "," -. Running that on my laptop I get the expected output:

src/Logger/GitHub/GitDiffFileProvider.php

[maks-rafalko]

I didn't dig yet, but I think this is related to how GitHub clones the repo.

Try locally to do the same as GitHub does, here:

/usr/bin/git -c protocol.version=2 fetch --no-tags --prune --progress --no-recurse-submodules --depth=1 origin +ec43076a8189aa8a0dfc6c8028a6e5c9bc178e21:refs/remotes/pull/1654/merge

then here

infection/.github/workflows/mt-annotations.yaml

Lines 50 to 51 in 4e21783

	git fetch --depth=1 origin $GITHUB_BASE_REF
	php bin/infection -j2 --git-diff-lines --git-diff-base=origin/$GITHUB_BASE_REF --logger-github --ignore-msi-with-no-mutations --only-covered

so, probably after such cloning, there won't be any diff by using git diff 'origin/master...HEAD' --diff-filter='AM' --name-only | grep src/ | paste -s -d "," -

[bdsl]

ah right, that depth=1 truncates the history. Without access to the history it would be impossible to distinguish between lines changed on the head branch and lines changed on master.

[bdsl]

I think we can make it use the merge-base for comparison if available in the checkout, if not we can fall back to doing a direct diff between commits. Probably worth advising users to make CI servers do deep checkouts to ensure we can find the merge-base and don't mutate more (or occasionally less) than needed.

[maks-rafalko]

👍

cool, thank you @bdsl

[bdsl]

@maks-rafalko Welcome, thank you too. Do you know when this is most likely to be in a release?

[bdsl]

It would be great if it can be on the 0.26 branch (maybe requires backport / cherry-pick ?) since we're still using PHP 7.4

[maks-rafalko]

if you want it to be released for 7.4, please open a new PR with cherry-picking it to 0.26 branch as a target and I will release it ASAP

[bdsl]

Will do, thanks.

[bdsl]

PR for 0.26: #1655