Wow, that is fantastic. Thank you so much for collecting the data Cosimo. I too would agree that an average savings of 548 bytes would not be sufficient savings to justify the increased complexity and likely increased decode times. On Thu, May 21, 2015 at 12:53 PM, Cosimo Lupo <cosimo.lupo@daltonmaag.com> wrote: > I just run a test on all the Dalton Maag Library fonts, as well as on a > bunch of popular Google Fonts. The average saving per font is just... 548 > bytes. > Interestingly, this average value is almost exactly the same for both DaMa > fonts and Google Fonts (read below for the full stats). > > Note that the average number of glyph per font ranged about 700 and > something. Given the lsb is stored as un unsigned short, one would expect > an average saving of 1400 bytes, but Brotli entropy coding seems to save a > little less when lsb are not there. > > Do you think the overall trade-off between compression gain and > reconstruction cost is worth it? > I’m not very impressed, I must confess. I was hoping for a bit more... > > Cheers, > > C. > > > ### > > Dalton Maag Library > > number of fonts: 164 > average number of glyphs per font: 732 > total size with hmtx: 8857900 > average size with hmtx: 54011 > total size without hmtx: 8767928 > average size without hmtx: 53462 > total difference: 89972 > average difference: 548 > > > "Top 50” Google Fonts from Skyfonts > > number of fonts: 187 * > average number of glyphs per font: 708 > total size with hmtx: 10758700 > average size with hmtx: 57533 > total size without hmtx: 10656396 > average size without hmtx: 56986 > total difference: 102304 > average difference: 547 > > > * (the number of fonts was 221 fonts, however 34 of them had different > xMin and lsb, so I excluded those from the average) > > > >
Received on Thursday, 21 May 2015 20:04:46 UTC