[CodeExtractor] Support merge commits

When using cherry-pick, it wouldn't handle merges properly, since it
requires a `-m 1` argument for it to work properly.

By switching to using `git rebase --root --onto`, we can take the branch
that now is a "orphan" branch, and apply the commits onto the target
remote's target branch.

Left extra code in there to show other possibilities (since this isn't
getting merged in away)

lib/code_extractor.rb

@@ -94,6 +94,16 @@ def extract_commits extractions, upstream_name
       end
     end
 
+    # Three step process to filter out the commits we want in three passes:
+    #
+    #   - Move code we want to keep into a separate tmp dir (using @prune_script)
+    #   - Prune anything that isn't in that subdirectory
+    #   - Move the tmp directory back into the root of the directory
+    #
+    #
+    # Note:  We don't use `--subdirectory-filter` here as it will remove merge
+    # commits, which we don't want.
+    #
     def prune_commits extractions
       puts "Pruning commits…"
 
@@ -102,7 +112,11 @@ def prune_commits extractions
       Dir.chdir git_dir do
         `git checkout -b #{prune_branch} #{@source_branch}`
         `git filter-branch -f --prune-empty --tree-filter #{@prune_script} HEAD`
-        `git filter-branch -f --prune-empty --subdirectory-filter #{@keep_directory}`
+        `git filter-branch -f --prune-empty --index-filter '
+            git read-tree --empty
+            git reset $GIT_COMMIT -- #{@keep_directory}
+          'HEAD`
+        `git filter-branch -f --prune-empty --tree-filter 'mv #{@keep_directory}/* .' HEAD`
       end
     end
 
@@ -188,8 +202,28 @@ def inject_commits target_base_branch, upstream_name
           echo "(transferred from #{upstream_name}@$GIT_COMMIT)"
         ' -- #{prune_branch}`
 
-        `git checkout --no-track -b #{inject_branch} #{@reference_target_branch}`
-        `git cherry-pick ..#{prune_branch}`
+        # Old (bad:  doesn't handle merges)
+        #
+        # `git checkout --no-track -b #{inject_branch} #{@reference_target_branch}`
+        # `git cherry-pick ..#{prune_branch}`
+        #
+        #
+        # Attempted #1 (not working, but uses older `git` methods)
+        #
+        # `git checkout --orphan #{inject_branch}`
+        # `git commit -m "Dummy Init commit"`
+        # orphan_commit = `git log --pretty="%H" -n1`.chomp
+        # prune_first   = `git log --pretty="%H" --reverse -n1 #{prune_branch}`.chomp
+        # `git rebase --onto #{orphan_commit} #{prune_first} #{inject_branch}`
+        # `git replace #{orphan_commit} #{@reference_target_branch}`
+        #
+        #
+        # Better
+        #
+        # Ref:  git rebase --onto code_extractor_inject_the_extracted --root
+        #
+        `git checkout --no-track -b #{inject_branch} #{prune_branch}`
+        `git rebase --preserve-merges --root --onto #{@reference_target_branch} #{inject_branch}`
       end
     end
 

test/code_extractor_reinsert_test.rb

@@ -34,6 +34,8 @@ def test_unextract_an_extraction
       ]
 
       refute Dir.exist?  "foo"
+      refute File.exist?  "README.md"
+
       assert File.exist? "qux"
       assert File.exist? "lib/foo/bar"
       assert File.exist? "lib/foo/baz"
@@ -129,6 +131,68 @@ def test_reinsert_when_code_extractor_was_not_used
       ]
 
       refute Dir.exist?  "foo"
+      refute File.exist?  "README.md"
+
+      assert File.exist? "qux"
+      assert File.exist? "lib/foo/bar"
+      assert File.exist? "lib/foo/baz"
+    end
+  end
+
+  # Basically, a test to ensure we are using a rebase and not a cherry-pick
+  # method, since `git cherry-pick` doesn't handle merge commits well.
+  def test_reinsert_when_there_is_a_merge_commit
+    # original extraction to work off of, in which we "un-extract" this later
+    create_base_repo
+    set_extractions ["foo"]
+    run_extraction
+
+    # Perform updates to extracted repo to simulate changes since extraction
+    perform_merges_of_extracted_code
+    apply_new_commits_on_extracted_repo do
+      checkout_b 'master', 'origin/master'
+
+      update_file "foo/bar", "Updated Bar Content"
+      commit "update bar content"
+
+      checkout_b 'add_baz', 'master'
+
+      update_file "foo/baz", "Baz Content"
+      commit "add new baz"
+
+      checkout 'master'
+      merge 'add_baz'
+
+      add_file "README.md", "READ ME!"
+      commit "add README"
+    end
+    update_extraction_hash
+
+    # Run new extraction, with some extra commits added to the new repo that
+    # has been extracted previously
+    #
+    # This next line will run the actual extraction we are testing
+    #
+    # aka:  updated commits and puts 'lib/foo' back into the original repo
+    run_extraction
+
+    in_git_dir do
+      assert_commits [
+        "Move foo/ into lib/",
+        "Merged branch 'add_baz' into master",
+        "add new baz",
+        "update bar content",
+        "Re-insert extractions from MyOrg/extracted_repo",
+        "Merged branch 'extract_my_extractions' into master",
+        "Extract my_extractions",
+        "Commit #3",
+        "add Bar content",
+        "Initial Commit"
+      ]
+
+      refute Dir.exist?  "foo"
+      refute File.exist?  "README.md"
+
       assert File.exist? "qux"
       assert File.exist? "lib/foo/bar"
       assert File.exist? "lib/foo/baz"
@@ -170,6 +234,9 @@ def apply_new_commits_on_extracted_repo &block
 
         update_file "foo/baz", "Baz Content"
         commit "add new baz"
+
+        add_file "README.md", "READ ME!"
+        commit "add README"
       end
     end