fix cache updates for scm aware queries

Summary:
A negative logic bug meant that we'd never update our remotename->hash mapping.

This could lead to a feedback loop; each attempt to `hg log` would cause hg to update blackbox.log
and thus cause subscriptions to be re-evaluated and thus run `hg log` again.

Reviewed By: simpkins

Differential Revision: D8337997

fbshipit-source-id: b49a717af752264bc3de8741fc4a2040d39cbfa2

scm/Mercurial.cpp

@@ -157,7 +157,7 @@ w_string Mercurial::mergeBaseWith(w_string_piece commitId, w_string requestId)
     // we collected because we know that a pending change will advise clients of
     // the new state
     auto cache = cache_.wlock();
-    if (!fileTimeEqual(startDirState, cache->dirstate)) {
+    if (fileTimeEqual(startDirState, cache->dirstate)) {
       cache->mergeBases[idString] = outputs.first;
     }
   }

tests/integration/test_scm.py

@@ -504,6 +504,21 @@ def test_scmHg(self):
         )
         self.assertFileListsEqual(res["files"], ["a/b/c/f1"])
 
+        # Sanity check that the mergebase cache is basically working.
+        # Ideally we'd check for a single miss in the server instance log, but
+        # since we may run variations on this same test against the same instance,
+        # it is hard to guarantee that the count would be meaningful here.  So
+        # instead we look at the volume of log calls and check that it is below
+        # a reasonable threshold.
+        blackbox = os.path.join(root, ".hg", "blackbox.log")
+        if os.path.exists(blackbox):
+            misses = 0
+            with open(blackbox) as f:
+                for line in f:
+                    if "'ancestor(.,TheMaster)' exited" in line:
+                        misses += 1
+            self.assertLessEqual(misses, 20)
+
     def getConsolidatedFileList(self, dat):
         fset = set()
         for _ in dat: