fb55 / entities / 17417266985 / 1

84%

master: 83%

Pull #1944

github

Improve decode trie size

Reduced length of the encoded trie by 21% for HTML (from 15,242 words (uint16) to 12075), by:
- Adding a semicolon flag on value nodes, which means semicolons don't have to be encoded anymore.
- Compacting dict keys: two keys are now stored in each word, instead of one
- Compacting runs of 3 or more characters: successive characters are now also stored as two characters per word

Also reduces the size of the encoded table by 32% (after the previous change) by encoding it as base64.

260 of 285 branches covered (91.23%)

Branch coverage included in aggregate %.

984 of 1219 relevant lines covered (80.72%)

290427.78 hits per line