I have one (probably quite naive) implementation here:

https://github.com/djc/runa/blob/master/runac/parser.py#L54

(Runa uses tabs for indentation, not spaces, but it's basically the same thing.) Whitespace at the start of a line is a separate token, then I add a thin layer of processing that turns a change of indentation level (i.e. count chars in the token contents) into INDENT or DEDENT tokens.