fix(lfs): ensure that LFS objects are checked out

[sxlijin]

The current implementation of lfs: true does not actually guarantee that LFS objects are checked out:

• it first does a git lfs fetch origin (which does not update the working copy),
• and then relies on git checkout in combination with the smudge filter to check LFS objects out.

However, there are numerous reports of flakiness with this approach (and we've seen this ourselves in our own usage); see #270.

Adding a git lfs checkout should make this more robust.

Unfortunately I don't have a consistent repro for this issue, beyond the anecdata from our own workflow failures.

[ceisen1]

Any particular reason why you closed this PR @sxlijin?

[sxlijin]

Hmm... I thought we had internally come to a different conclusion about this, but double checking our CI setup we didn't: we always run git lfs checkout after actions/checkout.

One of the folks on my team had the hypothesis that all the flakiness issues people have reported have to do with persistent/shared state, since we use persistent self-hosted runners.

[ceisen1]

Got it. That makes sense based on your description. But I do believe your original comments are correct that only running fetch is never going to update the lfs files in the working copy.

My one suggestion for your PR would be to move the git lfs checkout call into an if block where settings.lfs is True

[Ravio1i]

Hey there, we do face the same issue, on windows we have to run

   - uses: actions/checkout@v3
      with:
        lfs: true

    - name: Git LFS Pull 
      run: git lfs pull

to ensure it actually gets pulled

[bdeeming]

Consider

Suggested change

	core.startGroup('Checking out LFS objects')
	await git.lfsCheckout()
	core.endGroup()
	if (settings.lfs && !settings.sparseCheckout) {
	core.startGroup('Checking out LFS objects')
	await git.lfsCheckout()
	core.endGroup()
	}