Hi, I just pushed the packed trie implementation from http://flpsed.org/hgweb/dillo_hyphen_mem to the main repository. It considerably reduces the memory consumption of the hyphenator. The <lang>.pat files of the original implementation without any changes. However one can reduce memory consumption even further by precompressing the patterns using the new test/trie program: Usage: trie <pattern file> > <prefix>/lib/dillo/hyphenation/<lang>.pat.trie e.g: ./trie hyph-de-1996.pat > /usr/local/lib/dillo/hyphenation/de.pat.trie If the <lang>.pat.trie is found, dillo will use this precompressed trie instead of the <lang>.pat file. These changes should not have any effect on where the hyphenator breaks up words. If you find any regressions or changes, please report them. Cheers, Johannes