For a long time (since 2019) git-annex has been progressively converting from FilePath to RawFilePath (aka ByteString).

The reason is mostly performance, also a simpler representation of filepaths that doesn't need encoding hacks to support non-UTF8 values.

Some commands like git-annex find use RawFilePath end-to-end. But this conversion is not yet complete. This is a todo to keep track of the status.

  • The Abstract FilePath proposal (AFPP) has been implemented, and so a number of libraries like unix and directory now have versions that operate on OSPath. That could be used in git-annex eg for things like getDirectoryContents, when built against those versions. (But OSPath uses ShortByteString, while RawFilePath is ByteString, so conversion still entails a copy.)
  • withFile remains to be converted, and is used in several important code paths, including Annex.Journal and Annex.Link. There is a RawFilePath version in file-io library, but that is not currently a git-annex dependency. (withFile is in base, and base is unlikely to convert to AFPP soon)

The following patch can be useful to find points where conversions are done. Especially useful to identify cases where a value is converted FilePath -> RawFilePath -> FilePath.

diff --git a/Utility/FileSystemEncoding.hs b/Utility/FileSystemEncoding.hs
index 2a1dc81bc1..03e6986f6e 100644
--- a/Utility/FileSystemEncoding.hs
+++ b/Utility/FileSystemEncoding.hs
@@ -84,6 +84,9 @@ encodeBL = L.fromStrict . encodeBS
 encodeBL = L8.fromString
 #endif

+debugConversions :: String -> IO ()
+debugConversions s = hPutStrLn stderr ("conversion: " ++ s)
+
 decodeBS :: S.ByteString -> FilePath
 #ifndef mingw32_HOST_OS
 -- This does the same thing as System.FilePath.ByteString.decodeFilePath,
@@ -92,6 +95,7 @@ decodeBS :: S.ByteString -> FilePath
 -- something other than a unix filepath.
 {-# NOINLINE decodeBS #-}
 decodeBS b = unsafePerformIO $ do
+   debugConversions (show b)
    enc <- Encoding.getFileSystemEncoding
    S.useAsCStringLen b (GHC.peekCStringLen enc)
 #else
@@ -106,17 +110,19 @@ encodeBS :: FilePath -> S.ByteString
 -- something other than a unix filepath.
 {-# NOINLINE encodeBS #-}
 encodeBS f = unsafePerformIO $ do
+   debugConversions f
    enc <- Encoding.getFileSystemEncoding
-   GHC.newCStringLen enc f >>= unsafePackMallocCStringLen
+   b <- GHC.newCStringLen enc f >>= unsafePackMallocCStringLen
+   return b
 #else
 encodeBS = S8.fromString
 #endif

 fromRawFilePath :: RawFilePath -> FilePath
-fromRawFilePath = decodeFilePath
+fromRawFilePath = decodeBS -- decodeFilePath

 toRawFilePath :: FilePath -> RawFilePath
-toRawFilePath = encodeFilePath
+toRawFilePath = encodeBS -- encodeFilePath

 {- Truncates a FilePath to the given number of bytes (or less),
  - as represented on disk.