localrepo: silently recover from abandoned transaction

Summary:
As we changed how Ctrl+C is handled, there would be more "abandoned
transaction"s, just silently cleaned them up.

Practically, we should never hit the "abandoned transaction found" case,
because the `journal` file only contains how to truncate files to recover. Only
legacy revlog uses file truncation and we don't use them in production.

There might be a `journal.backupfiles` that tracks what files to rename to
restore backup. Most of the key files like `bookmarks` are in metalog therefore
no need to use backupfiles to restore. The remaining ones also seem fine to not
restore in the old way:
- dirstate: we use atomic replace on transaction close so it seems unnecessary
  to backup
- fncache: not used by remotefilelog
- phaseroots: not used when narrow-heads is on
- commitcloudstate.<name>.<hash>: source of truth moved to metalog by D35232804 (3807e9662b).
- commitcloud/backedupheads.remote: used for "backed up" rendering in smartlog,
  not fatal if not restored.

We might clean them up later. For now, let's just silently recover without
printing the legacy messages.

Note D33482320 (e75c4ca2d8) is incorrect about `journal` being empty = no-op recover,
since `journal.backupfiles` could be non-empty and trigger side effects on
`recover`.

Reviewed By: muirdm

Differential Revision: D42912639

fbshipit-source-id: 8a576d2b73b2ffc053e4fc699f323a1f72498feb

eden/scm/edenscm/localrepo.py

@@ -1698,15 +1698,7 @@ class localrepository(object):
             # No existing transaction - this is the normal case.
             pass
         else:
-            if stat.st_size > 0:
-                # Non-empty transaction already exists - bail.
-                raise errormod.AbandonedTransactionFoundError(
-                    _("abandoned transaction found"),
-                    hint=_("run '@prog@ recover' to clean up transaction"),
-                )
-            else:
-                self.ui.status(_("cleaning up empty abandoned transaction\n"))
-                self.recover()
+            self.recover()
 
         idbase = b"%.40f#%f" % (random.random(), time.time())
         ha = hex(hashlib.sha1(idbase).digest())
@@ -1959,7 +1951,7 @@ class localrepository(object):
     def recover(self):
         with self.lock():
             if self.svfs.exists("journal"):
-                self.ui.status(_("rolling back interrupted transaction\n"))
+                self.ui.debug("rolling back interrupted transaction\n")
                 vfsmap = {
                     "": self.svfs,
                     "shared": self.sharedvfs,

eden/scm/tests/test-journal-exists.t

@@ -10,20 +10,19 @@
   $ newclientrepo foo test:repo_server book
   $ cd ../repo
 
+Journal is cleaned up automatically.
   $ echo something > .hg/store/journal
 
   $ echo foo > a
   $ hg ci -Am0
-  abort: abandoned transaction found!
-  (run 'hg recover' to clean up transaction)
-  [255]
+  couldn't read journal entry 'something\n'!
 
   $ hg recover
-  rolling back interrupted transaction
-  couldn't read journal entry 'something\n'!
+  no interrupted transaction available
+  [1]
 
 Empty journal is cleaned up automatically.
   $ touch .hg/store/journal
   $ hg ci -Am0
-  cleaning up empty abandoned transaction
-  rolling back interrupted transaction
+  nothing changed
+  [1]