work around box.com webdav rename bug

Apparently box.com renaming is just buggy. I tried a couple of fixes:

* In case the http Manager was opening multiple connections and reaching
  different backend servers, I tried limiting the number of connections
  to 1. Didn't help.
* To make sure it was not a http connection reuse problem, I tried
  rewriting how exportAction works, so that the same http connection
  is clearly open. Didn't help.

So, disable renaming of exports for box.com. It would be good to test it
with some other webdav server.

This commit was sponsored by John Peloquin on Patreon.

2 files changed, 16 insertions, 13 deletions

diff --git a/Remote/WebDAV.hs b/Remote/WebDAV.hs
index 5e853ae22..8c72365e6 100644
--- a/Remote/WebDAV.hs
+++ b/Remote/WebDAV.hs
@@ -201,9 +201,15 @@ checkPresentExportDav r mh _k loc = case mh of
 		either giveup return v
 
 renameExportDav :: Maybe DavHandle -> Key -> ExportLocation -> ExportLocation -> Annex Bool
-renameExportDav mh _k src dest = runExport mh $ \dav -> do
-	moveDAV (baseURL dav) (exportLocation src) (exportLocation dest)
-	return True
+renameExportDav Nothing _ _ _ = return False
+renameExportDav (Just h) _k src dest
+	-- box.com's DAV endpoint has buggy handling of renames,
+	-- so avoid renaming when using it.
+	| boxComUrl `isPrefixOf` baseURL h = return False
+	| otherwise = runExport (Just h) $ \dav -> do
+		maybe noop (void . mkColRecursive) (locationParent (exportLocation dest))
+		moveDAV (baseURL dav) (exportLocation src) (exportLocation dest)
+		return True
 
 runExport :: Maybe DavHandle -> (DavHandle -> DAVT IO Bool) -> Annex Bool
 runExport Nothing _ = return False
@@ -213,7 +219,10 @@ configUrl :: Remote -> Maybe URLString
 configUrl r = fixup <$> M.lookup "url" (config r)
   where
 	-- box.com DAV url changed
-	fixup = replace "https://www.box.com/dav/" "https://dav.box.com/dav/"
+	fixup = replace "https://www.box.com/dav/" boxComUrl
+
+boxComUrl :: URLString
+boxComUrl = "https://dav.box.com/dav/"
 
 type DavUser = B8.ByteString
 type DavPass = B8.ByteString
diff --git a/doc/todo/export.mdwn b/doc/todo/export.mdwn
index 45fc56995..efa8b1c38 100644
--- a/doc/todo/export.mdwn
+++ b/doc/todo/export.mdwn
@@ -40,12 +40,6 @@ Low priority:
   Run each pair in turn. Then run the current rename code. Although this
   still probably misses cases, where eg, content cycles amoung 3 files, and
   the same content amoung 3 other files. Is there a general algorythm?
-* Exporting to box.com via webdav, a rename of a file behaves
-  oddly. The rename to the temp file succeeds, but the rename of the temp
-  file to the final name fails.
-  Also, sometimes the delete of the temp file that's done as a fallback
-  fails to actually delete it.
-  Hypothesis: Those are done in separate http connections and it might be
-  talking to two different backend servers that are out of sync.
-  So, making export cache connections might help. Update: No, caching
-  connections did not solve it.
+* webdav: When a file in a subdirectory gets deleted,
+  the webdav collection is remains, empty. Need to check if collection is
+  empty, and delete otherwise.