fix memory leak

filterM is not a good idea if you were streaming in a large list of files.

Fixing this memory leak that I introduced earlier today was a PITA because
to avoid the filterM, it's necessary to do the filtering only after
building up the data structures like BackendFile, and that means each
separate data structure needs it own function to apply the filter,
at least in this naive implementation.

There is also a minor performance regression, when using copy/drop/get/fsck
with a filter, git is now asked to look up attributes for all files,
since that now comes before the filter is applied. This is only a very
minor thing, since getting the attributes is very fast and --exclude was
probably not typically used to speed it up.

1 files changed, 7 insertions, 8 deletions

diff --git a/Limit.hs b/Limit.hs
index 51f3fc950..602da7001 100644
--- a/Limit.hs
+++ b/Limit.hs
@@ -22,20 +22,19 @@ import Utility
 
 type Limit = Utility.Matcher.Token (FilePath -> Annex Bool)
 
-{- Filter out files not matching user-specified limits. -}
-filterFiles :: [FilePath] -> Annex [FilePath]
-filterFiles l = do
-	matcher <- getMatcher
-	filterM (Utility.Matcher.matchM matcher) l
-
 {- Checks if there are user-specified limits. -}
 limited :: Annex Bool
-limited = (not . Utility.Matcher.matchesAny) <$> getMatcher
+limited = (not . Utility.Matcher.matchesAny) <$> getMatcher'
 
 {- Gets a matcher for the user-specified limits. The matcher is cached for
  - speed; once it's obtained the user-specified limits can't change. -}
-getMatcher :: Annex (Utility.Matcher.Matcher (FilePath -> Annex Bool))
+getMatcher :: Annex (FilePath -> Annex Bool)
 getMatcher = do
+	m <- getMatcher'
+	return $ Utility.Matcher.matchM m
+
+getMatcher' :: Annex (Utility.Matcher.Matcher (FilePath -> Annex Bool))
+getMatcher' = do
 	m <- Annex.getState Annex.limit
 	case m of
 		Right r -> return r