Split out !important parsing into a separate function.
Support for dumping bytecode to a file handle in some kind of human-readable format.
Strings can be represented in the bytecode as a pointer, length pair rather than embedding the string data into the bytecode -- all strings are interned by the core syntax parser.
Add todo relating to early destruction of parser object (it shouldn't be needed once parsing is complete). Note documented issue surrounding interned string dictionary, however.
In general, it seems wasteful to create a new dictionary containing string representations of keywords for every single parser instance. It would be better to have one central (statically allocated?) dictionary for this and then each parser instance can have a smaller dictionary containing any unknown strings contained within the stylesheet being parsed (e.g. string constants or URLs).
svn path=/trunk/libcss/; revision=5627