Facebook snapshot:
- minified, decompressed: 42Mb;
- minifed, brotli-compressed (max compression): 9Mb;
- entropy
- non-optimized dictionary, decompressed: 13Mb;
- non-optimized dictionary, compressed with brotli: 1.9Mb;
- minified, entropy-compressed: 6.9Mb
Two notes about the non-optimized dictionary:
- it contains lots of useless data, we should be able to reduce it a lot;
- the same dictionary should work across many releases of Facebook.
Cool. So am I interpreting this right, a current comparison would be minified, brotli-compressed: 9 MB to minified, entropy encoded 6.9 MB + dictionary, brotli 1.9 MB = 8.8 MB? That's splendid news.
I had a quick question for you: We churn strings relatively quickly so we will probably need strings which don't appear in the dictionary in the files if we're to reuse dictionaries. Does the format have an escape hatch for that?