merge: flush any deferred writes before, and after, running any workers

Since we fork to create workers, any changes they queue up will be lost after
the worker terminates, so the easiest solution is to have each worker flush
the writes they accumulate--we are close to the end of the merge in any case.

To prevent duplicated writes, we also have the master processs flush before
forking.

In an in-memory merge (M2), we'll instead disable the use of workers.

Differential Revision: https://phab.mercurial-scm.org/D628

mercurial/merge.py

@@ -1126,6 +1126,10 b' def batchremove(repo, wctx, actions):'
                            "(consider changing to repo root: %s)\n") %
                          repo.root)
 
+    # It's necessary to flush here in case we're inside a worker fork and will
+    # quit after this function.
+    wctx.flushall()
+
 def batchget(repo, mctx, wctx, actions):
     """apply gets to the working directory
 
@@ -1161,6 +1165,10 b' def batchget(repo, mctx, wctx, actions):'
     if i > 0:
         yield i, f
 
+    # It's necessary to flush here in case we're inside a worker fork and will
+    # quit after this function.
+    wctx.flushall()
+
 def applyupdates(repo, actions, wctx, mctx, overwrite, labels=None):
     """apply the merge action list to the working directory
 
@@ -1229,6 +1237,10 b' def applyupdates(repo, actions, wctx, mc'
         progress(_updating, z, item=item, total=numupdates, unit=_files)
     removed = len(actions['r'])
 
+    # We should flush before forking into worker processes, since those workers
+    # flush when they complete, and we don't want to duplicate work.
+    wctx.flushall()
+
     # get in parallel
     prog = worker.worker(repo.ui, 0.001, batchget, (repo, mctx, wctx),
                          actions['g'])