work around minimum part size problem

When uploading the last part of a file, which was 640229 bytes, S3 rejected
that part: "Your proposed upload is smaller than the minimum allowed size"

I don't know what the minimum is, but the fix is just to include the last
part into the previous part. Since this can result in a part that's
double-sized, use half-sized parts normally.

Remote/S3.hs

@@ -181,9 +181,16 @@ store r h = fileStorer $ \k f p -> do
 				}
 		uploadid <- S3.imurUploadId <$> sendS3Handle h startreq
 
-		-- The actual part size will be a even multiple of the
+		{- The actual part size will be a even multiple of the
-		-- 32k chunk size that hGetUntilMetered uses.
+		 - 32k chunk size that hGetUntilMetered uses.
-		let partsz' = (partsz `div` toInteger defaultChunkSize) * toInteger defaultChunkSize
+		 -
+		 - Also, half-size parts are used. This is so that
+		 - the final part of a file can be rolled into the
+		 - last full-size part, which avoids a problem when the
+		 - final part could otherwise be too small for S3 to accept
+		 - it.
+		 -}
+		let partsz' = (partsz `div` toInteger defaultChunkSize `div` 2) * toInteger defaultChunkSize
 
 		-- Send parts of the file, taking care to stream each part
 		-- w/o buffering in memory, since the parts can be large.
@@ -195,7 +202,7 @@ store r h = fileStorer $ \k f p -> do
 					else do
 						-- Calculate size of part that will
 						-- be read.
-						let sz = if fsz - pos < partsz'
+						let sz = if fsz - pos < partsz' * 2
 							then fsz - pos
 							else partsz'
 						let p' = offsetMeterUpdate p (toBytesProcessed pos)

doc/special_remotes/S3.mdwn

@@ -21,10 +21,14 @@ the S3 remote.
 * `chunk` - Enables [[chunking]] when storing large files.
   `chunk=1MiB` is a good starting point for chunking.
 
-* `partsize` - Specifies the largest object to attempt to store in the
+* `partsize` - Amazon S3 only accepts uploads up to a certian file size,
-  bucket. Multipart uploads will be used when storing larger objects.
+  and storing larger files requires a multipart upload process.
-  This is not enabled by default, but can be enabled or changed at any
+  Setting `partsize=1GiB` is recommended for Amazon S3; this will
-  time. Setting `partsize=1GiB` is reasonable for S3.
+  cause multipart uploads to be done using parts up to 1GiB in size.
+
+  This is not enabled by default, since other S3 implementations may
+  not support multipart uploads, but can be enabled or changed at any
+  time.
 
 * `keyid` - Specifies the gpg key to use for [[encryption]].