This code takes a precomputed LZW dictionary and uses it to encode or decode.
Why?
This is useful for transmitting LZW-compressed buffers between endpoints that you want to always start from the same state- a good, non-empty dictionary. LZW normally takes a lot of input (probably several K at least) to build up good dictionary content. If you are transmitting a lot of small buffers it would not make sense to LZW-encode them individually. On the other hand, if you LZW encode the stream of buffers, and sometime later need to restart the sender or receiver, you'd need to restart both of them (or have the one quiescently digest the stream from the beginning) to re-sync their dictionaries.
Or, if you have a third system that needs to decode some content from the message stream between the first two, it would be necessary to know the dictionary state (as well as a proper boundary for the bitcodes in the encoded stream). Both of these requirements can be met by using a fixed LZW dictionary and a fixed bitcode length for the encoded indexes.
First, generate a good LZW dictionary on a suitably-large data sample. Here we generate a dictionary with up to one million sequences.
% ./mlzw -e -i census-names.in -o census-names.out -C dict -D 1000000
Now the dictionary file can be used to encode and decode:
% ./mlzw -e -i names.in -o names.out -c dict
% ./mlzw -d -i names.out -i names.org -c dict
The mlzw command is demonstrating the underlying C API which is the real purpose of mini-lzw.
mlzw_load(lzw, "names.lzw");
rc = mlzw_recode(mode, lzw, input, input_len, output, &output_len);
if (rc < 0) ...