dvc: introduce merge-driver

[efiop]

Related to #4162

• ❗ I have followed the Contributing to DVC checklist.

• 📖 If this PR requires documentation updates, I have created a separate PR (or issue, at least) in dvc.org and linked it here.

Thank you for the contribution - we'll try to review it as soon as possible. 🙏

TODO:

• more tests

[efiop]

These stubs are for the future, as merging dvcfiles is potentially a handy thing to have.

[efiop]

We have a similar thing in _collect_dir in BaseTree's, but these will be unified as a part of #4144 , once dir hash computation is detached from cache.

[pared]

@efiop
What about conflicting files? I prepared sample script:

#!/bin/bash

rm -rf repo
mkdir repo

pushd repo

git init --quiet
dvc init --quiet

git add -A
git commit -am "init"

dvc install
echo "*.dvc merge=dvc" >> .gitattributes
git add -A
git commit -m "setup hooks"

git checkout -b one
mkdir data
echo foo >> data/foo

dvc add data
git add -A
git commit -m "add foo"

git checkout master

git checkout -b two
mkdir data
echo bar >> data/foo

dvc add data
git add -A
git commit -m "add foo but bar"

git checkout master

git merge one --no-edit
git merge two --no-edit
# I think we should get error now

Where I have 2 branches:
one that has dvc add-ed data/foo with foo content
two that has dvc add-ed data/foo with bar content

and then, on master I merge one and then two.
My data.dvc gets silently overridden so that data/foo contains bar - the two branch content.
Shouldn't we get some error?

[efiop]

@pared Sorry for the delay.

git checkout master
git merge one --no-edit
# I think we should get error now

Not really, because there is nothing in master, so there is no conflict.

git merge two --no-edit

And here you get a conflict, as expected.

[pared]

@efiop sorry, I meant that I would expect error after the comment.
And yes, there is conflict, but my .dvc file gets Auto-merged. It is some indicator of failure, but it is easy to omit. I would expect that git will not try to auto-merge. Choosing which file is the proper one should be users responsibility.
Now the content of data/foo depends on the order of merge:

git merge one --no-edit
git merge two --no-edit
# cat data/foo == bar

git merge two --no-edit
git merge one --no-edit
# cat data/foo == foo

If we were doing same thing (tracking data/foo) in git, we would get unresolved conflict.

EDIT:
git equivalent:

#!/bin/bash

rm -rf repo
mkdir repo

pushd repo

git init --quiet
echo other_file >> other_file
git add other_file

git add -A
git commit -am "init"

git checkout -b one
mkdir data
echo foo >> data/foo

git add data
git add -A
git commit -m "add foo"

git checkout master

git checkout -b two
mkdir data
echo bar >> data/foo

git add data
git add -A
git commit -m "add foo but bar"

git checkout master

git merge two --no-edit
git merge one --no-edit
#merge conflict

[efiop]

@pared Ah, got it. Yes, that's a bug, thank you! 🙏 Finally getting to creating additional tests for it, will be sure to add one for that as well.