[krokicki]

encoding-type=url means percent-encoding, but url_encode() replaced spaces with +, so a file named my file.txt was listed as key my+file.txt. Clients undo the encoding by unquoting (boto3) or with decodeURIComponent (Neuroglancer), and neither turns + back into a space — so a key read out of a listing pointed at a file that doesn't exist, and fetching it 404'd.

Any file or directory whose name contains a space was affected, which is why it turned up immediately on real data: browsing a Fileglancer data link in Neuroglancer showed the entries but couldn't open them.

quote(s, safe='/') now handles it, which also encodes a literal + as %2B and so tells the two apart. / is still left alone so keys keep their structure.

New test builds a target over a tree with spaced names, asserts the listing reports my%20file.txt and sub%20dir/, and then reads the key back through the app to prove it resolves. Full suite passes (81).

Also: pagination inside a virtual prefix

A listing whose prefix stopped inside the virtual_prefix emitted its one synthetic CommonPrefixes entry unconditionally, so max-keys=0 answered with KeyCount=1, and a client that passed a continuation token got the same entry again. The filesystem branch of the same method honours both, so the two disagreed:

prefix=abc&delimiter=/&max-keys=0     -> KeyCount=1 common=['abc123/']   # wrong
prefix=tests/&delimiter=/&max-keys=0  -> KeyCount=0 common=[]            # right

That entry is the whole listing and is never truncated, so there is nothing to page through: max-keys=0 now returns an empty listing, and a continuation token can only be stale, which walk_path already treats as nothing to report. Covered by a test over both cases.

@StephanPreibisch