I noticed that our implementation of storing the servers map was faulty. It didn't handle the case that the remote ID was an empty string.[...] The patch below fixes the problem, while preserving backwards compatibility to existing map files. While looking at the two functions, I noticed that mapid2entry() escapes the ID before adding it to the key, whereas entry2mapid() unescapes the key, then extracts the ID. That doesn't look right. It should first split at "-", then unescape the parts. Congwu, you touched that code last. Do you agree that there is something to be fixed here?

Also, any comments on the patch below?