From c2651d64bc8a384f8669879c051577ee61da27ed Mon Sep 17 00:00:00 2001
From: Joey Hess <joey@kitenet.net>
Date: Sat, 30 Oct 2010 23:19:33 -0400
Subject: [PATCH] Fixed memory leak; git-annex no longer reads the whole file
 list from git before starting, and will be much faster with large repos.

---
 GitRepo.hs       | 16 +++++++++++++++-
 debian/changelog |  2 ++
 2 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/GitRepo.hs b/GitRepo.hs
index c8ba77133c..b6670e2ebe 100644
--- a/GitRepo.hs
+++ b/GitRepo.hs
@@ -27,6 +27,7 @@ module GitRepo (
 	configTrue,
 	run,
 	pipeRead,
+	hPipeRead,
 	attributes,
 	remotes,
 	remotesAdd,
@@ -198,6 +199,17 @@ pipeRead repo params = assertLocal repo $ do
 		ret <- hGetContentsStrict h
 		return ret
 
+{- Like pipeRead, but does not read output strictly; recommended
+ - for git commands that produce a lot of output that will be processed
+ - lazily. 
+ -
+ - ONLY AFTER the string has been read completely, You must call either
+ - getProcessStatus or forceSuccess on the PipeHandle. Zombies will result
+ - otherwise.-}
+hPipeRead :: Repo -> [String] -> IO (PipeHandle, String)
+hPipeRead repo params = assertLocal repo $ do
+	pipeFrom "git" (gitCommandLine repo params)
+
 {- Passed a location, recursively scans for all files that
  - are checked into git at that location. -}
 inRepo :: Repo -> FilePath -> IO [FilePath]
@@ -221,7 +233,9 @@ stagedFiles repo location = pipeNullSplit repo
  - parameter), and splits it into a list of files. -}
 pipeNullSplit :: Repo -> [String] -> IO [FilePath]
 pipeNullSplit repo params = do
-	fs0 <- pipeRead repo params
+	-- XXX handle is left open, this is ok for git-annex, but may need
+	-- to be cleaned up for other uses.
+	(handle, fs0) <- hPipeRead repo params
 	return $ split0 fs0
 	where
 		split0 s = filter (not . null) $ split "\0" s
diff --git a/debian/changelog b/debian/changelog
index d6cc73e49e..70e9052231 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -5,6 +5,8 @@ git-annex (0.03) UNRELEASED; urgency=low
   * Fix SIGINT handling.
   * Fix handling of files with unusual characters in their name.
   * Support building with Debian stable's ghc.
+  * Fixed memory leak; git-annex no longer reads the whole file list
+    from git before starting, and will be much faster with large repos.
 
  -- Joey Hess <joeyh@debian.org>  Thu, 28 Oct 2010 13:46:59 -0400