move, copy: Sped up seeking for annexed files to operate on by a factor of nearly 2x.

CHANGELOG

@@ -8,6 +8,8 @@ git-annex (8.20200720.2) UNRELEASED; urgency=medium
     some weird inheriting of ssh FDs by sshd. Bug was introduced in
     git-annex version 7.20200202.7.
   * Fix a bug in find --branch in the previous version.
+  * move, copy: Sped up seeking for annexed files to operate on by a factor
+    of nearly 2x.
 
  -- Joey Hess <id@joeyh.name>  Tue, 21 Jul 2020 12:58:30 -0400
 

Command/Copy.hs

@@ -57,8 +57,11 @@ seek o = startConcurrency commandStages $ do
 	
 	seeker = AnnexedFileSeeker
 		{ startAction = start o
-		, checkContentPresent = Nothing
+		, checkContentPresent = case fromToOptions o of
-		, usesLocationLog = False
+			Right (FromRemote _) -> Just False
+			Right (ToRemote _) -> Just True
+			Left ToHere -> Just False
+		, usesLocationLog = True
 		}
 
 {- A copy is just a move that does not delete the source file.

Command/Move.hs

@@ -55,11 +55,6 @@ data RemoveWhen = RemoveSafe | RemoveNever
 
 seek :: MoveOptions -> CommandSeek
 seek o = startConcurrency stages $ do
-	let seeker = AnnexedFileSeeker
-		{ startAction = start (fromToOptions o) (removeWhen o)
-		, checkContentPresent = Nothing
-		, usesLocationLog = False
-		}
 	case batchOption o of
 		NoBatch -> withKeyOptions (keyOptions o) False seeker
 			(commandAction . startKey (fromToOptions o) (removeWhen o))
@@ -67,6 +62,14 @@ seek o = startConcurrency stages $ do
 			=<< workTreeItems ww (moveFiles o)
 		Batch fmt -> batchAnnexedFilesMatching fmt seeker
   where
+	seeker = AnnexedFileSeeker
+		{ startAction = start (fromToOptions o) (removeWhen o)
+		, checkContentPresent = case fromToOptions o of
+			Right (FromRemote _) -> Nothing
+			Right (ToRemote _) -> Just True
+			Left ToHere -> Nothing
+		, usesLocationLog = True
+		}
 	stages = case fromToOptions o of
 		Right (FromRemote _) -> downloadStages
 		Right (ToRemote _) -> commandStages
@@ -103,9 +106,8 @@ describeMoveAction _ = "move"
 toStart :: RemoveWhen -> AssociatedFile -> Key -> ActionItem -> Remote -> CommandStart
 toStart removewhen afile key ai dest = do
 	u <- getUUID
-	ishere <- inAnnex key
+	if u == Remote.uuid dest
-	if not ishere || u == Remote.uuid dest
+		then stop
-		then stop -- not here, so nothing to do
 		else toStart' dest removewhen afile key ai
 
 toStart' :: Remote -> RemoveWhen -> AssociatedFile -> Key -> ActionItem -> CommandStart
@@ -188,11 +190,8 @@ toPerform dest removewhen key afile fastcheck isthere =
 			return False
 
 fromStart :: RemoveWhen -> AssociatedFile -> Key -> ActionItem -> Remote -> CommandStart
-fromStart removewhen afile key ai src = case removewhen of
+fromStart removewhen afile key ai src = 
-	RemoveNever -> stopUnless (not <$> inAnnex key) go
+	stopUnless (fromOk src key) $
-	RemoveSafe -> go
-  where
-	go = stopUnless (fromOk src key) $
 		starting (describeMoveAction removewhen) (OnlyActionOn key ai) $
 			fromPerform src removewhen key afile
 
@@ -247,11 +246,8 @@ fromPerform src removewhen key afile = do
  - When moving, the content is removed from all the reachable remotes that
  - it can safely be removed from. -}
 toHereStart :: RemoveWhen -> AssociatedFile -> Key -> ActionItem -> CommandStart
-toHereStart removewhen afile key ai = case removewhen of
+toHereStart removewhen afile key ai = 
-	RemoveNever -> stopUnless (not <$> inAnnex key) go
+	startingNoMessage (OnlyActionOn key ai) $ do
-	RemoveSafe -> go
-  where
-	go = startingNoMessage (OnlyActionOn key ai) $ do
 		rs <- Remote.keyPossibilities key
 		forM_ rs $ \r ->
 			includeCommandAction $

doc/todo/faster_key_lookup_for_limits.mdwn

@@ -0,0 +1,9 @@
+As part of the work in [[precache_logs_for_speed_with_cat-file_--buffer]],
+key lookups are now done twice as fast as before.
+
+But, limits that look up keys still do a key lookup, before the key
+is looked up efficiently. Avoiding that would speed up --in etc, probably
+another 1.5x-2x speedup when such limits are used. What that optimisation
+needs is a way to tell if the current limit needs the key or not. If it
+does, then match on it after getting the key (and precaching the location
+log for limits that need that), otherwise before getting the key.

doc/todo/precache_logs_for_speed_with_cat-file_--buffer.mdwn

@@ -33,10 +33,13 @@ and precache them.
 > > > * `sync --content` 2x speedup!
 > > > * `fsck --fast` 1.5x speedup
 > > > * `whereis` 1.5x speedup
+> > > * `copy --to --fast` twenty-five percent or so speedup
+> > > * `copy --to` 2x speedup
+> > > * `copy --from` 2x speedup
 > > >
-> > > Still todo:
+> > > For copy benchmarks, note that both repos had all files.
 > > >
-> > > * move, copy, drop, and mirror were left not using the location log caching yet
+> > > [[done]]
 
 Another thing that the same cat-file --buffer approach could be used with
 is to cat the annex links. Git.LsFiles.inRepoDetails provides the Sha
@@ -52,10 +55,4 @@ Some calls to lookupKey remain, and the above could
 be used to remove them and make it faster. The ones in Annex.View and
 Command.Unused seem most likely to be able to be converted.
 
-Also, limits that look up keys still do a key lookup, before the key is
+See also [[faster_key_lookup_for_limits]]
-looked up efficiently. (Before these changes, the same key lookup was done
-2x too..) Avoiding that would speed up --in etc, probably another 1.5x-2x
-speedup when such limits are used. What that optimisation needs is a way to
-tell if the current limit needs the key or not. If it does, then match on
-it after getting the key (and precaching the location log for limits that
-need that), otherwise before getting the key.