Fix N+1 in `MergeRequest#merge_request_diff_for`

Previously, this would issue a query for each unique `diff_refs_or_sha`
passed. This was because we didn't want to load other MR diffs into memory, as
they had some very large columns.

Now they are actually very small, and it's more efficient to just load them all
at once and do the finding in Ruby.

1 files changed, 11 insertions, 0 deletions

diff --git a/spec/models/merge_request_spec.rb b/spec/models/merge_request_spec.rb
index ff5a6f63010..f73f44ca0ad 100644
--- a/spec/models/merge_request_spec.rb
+++ b/spec/models/merge_request_spec.rb
@@ -1961,6 +1961,17 @@ describe MergeRequest do
         expect(subject.merge_request_diff_for(merge_request_diff3.head_commit_sha)).to eq(merge_request_diff3)
       end
     end
+
+    it 'runs a single query on the initial call, and none afterwards' do
+      expect { subject.merge_request_diff_for(merge_request_diff1.diff_refs) }
+        .not_to exceed_query_limit(1)
+
+      expect { subject.merge_request_diff_for(merge_request_diff2.diff_refs) }
+        .not_to exceed_query_limit(0)
+
+      expect { subject.merge_request_diff_for(merge_request_diff3.head_commit_sha) }
+        .not_to exceed_query_limit(0)
+    end
   end
 
   describe '#version_params_for' do