[Monorepo fixup] Reset lld contents in the swift/master branch to mat…

[hyp]

…ch apple/stable/20190619

The monorepo conversion for the swift/master branch brought in contents in
untracked directory that landed after the apple/stable/20190619 branch was cut.

This commit fixes this content by resetting the sub-trees to match apple/stable/20190619.

Here's how this commit was contructed:

Step 1: reset the subtrees to match the apple/stable/20190619 state

repos="debuginfo-tests libclc libcxxabi libunwind lld llgo openmp parallel-libs polly pstl"

for repo in $repos
do
git rm -rf $repo > /dev/null
latest_commit=$(git rev-list HEAD -n1 -- $repo)
tree=$(git rev-parse origin/apple/stable/20190619:$repo)
git read-tree --prefix=$repo $tree
echo "LATEST COMMIT FOR $repo -p $latest_commit"
done

Step 2: construct an octopus merge commit with new subtrees

tree=$(git write-tree)
git commit-tree -p HEAD -p ... -m "[Monorepo fixup] ..." $tree

[dexonsmith]

@hyp , is lld the only thing that's wrong, or is it all of the untracked directories? If the latter, I'd rather fix them all at the same time.

[hyp]

@hyp , is lld the only thing that's wrong, or is it all of the untracked directories? If the latter, I'd rather fix them all at the same time.

There's probably more, I need to check.

[dexonsmith]

@hyp , is lld the only thing that's wrong, or is it all of the untracked directories? If the latter, I'd rather fix them all at the same time.

There's probably more, I need to check.

Yeah, I think the problem is that LLDB branched later, and implicitly brought in a ton of other content in the untracked directories.

[dexonsmith]

@hyp , is lld the only thing that's wrong, or is it all of the untracked directories? If the latter, I'd rather fix them all at the same time.

There's probably more, I need to check.

Yeah, I think the problem is that LLDB branched later, and implicitly brought in a ton of other content in the untracked directories.

You can get the content from apple/stable/20190619/start, although I don't see that tag on llvm-project. Did you forward the tags yet from llvm-project-v5?

[hyp]

@dexonsmith Fixed up.

[dexonsmith]

@dexonsmith Fixed up.

I checked the tree and confirm it looks right. I'm not sure about having all those parents; it seems cleaner to just have a simple merge commit to the most recently already-merged apple/stable/20190619, which has the same tree.

[dexonsmith]

@swift-ci please test

[hyp]

@swift-ci please test

[dexonsmith]

@hyp we'll have to ignore the test failure:

17:26:09 + python /Users/buildnode/jenkins/workspace/apple-llvm-project-pr-macos/swift-ci/jobs/pull-request/check_changed_files.py
17:26:09 ERROR: File debuginfo-tests/nrvo-string.cpp is not allowed to be changed directly on swift/master

Let me know what you think about parents. I could be convinced that this giant octopus is the right thing, but it adds a lot of noise to git log --graph so I still think just re-merging from apple/stable/20190619 sets a better balance.

[dexonsmith]

I put up my proposed alternative as #16

[hyp]

Closing as #16 did it.