It's been hypothesized that using the glyf and loca table transformations
defined as part of WOFF2 format on the font file before subsetting and
patching may reduce the overall bytes transferred by the patch subset
method.
I was curious so I decided to do some quick testing to see how much of a
reduction it could produce. I added the woff2 glyf/loca transform to the
patch subset client code and then re-ran the CJK simulations. Using the
previous Patch Subset results as a baseline I generated the % byte and cost
reductions:
[image: comparison.woff2.cjk.sampled_5000.bytes.png]
[image: comparison.woff2.cjk.sampled_5000.cost.png]
The results are really good! For bytes transferred there was a median
reduction of 11% with reductions ranging from 0 to 15%. Which means in all
cases the transformation produces a net reduction in bytes. As expected the
costs also show similar reductions as a result of sending fewer bytes.
So far I've only got results for CJK. I'm still in the process of
re-running the LGC, Arabic and Indic simulations.
