|
From: Johan W. <joh...@te...> - 2006-02-25 15:51:09
|
On Sat, 25 Feb 2006 02:10:29 +0100, George Williams <gw...@si...> wrote: > However. No one seems to share my opinion. So the cvs tree now contains > a conversion algorithm which tries bisecting, trisecting, quadrasecting > ... a cubic curve until it finds a good quadratic fit. Well, I think I personally like the previous algorithm more (especially after the patch that followed my latest request - thank you for that one!). However, since I normally work with autotraced material, I have no big interest in the symmetric behaviour, since I can hardly expect the traced glyphs to be symmetric to begin with. If the cubic splines are hand made, I can see the merits of the new algorithm, and can relate to the points Alexey are making. Anyhow, the main reason I think the previous algorithm was better, is because it produced fewer points -- not because it produces smaller files (that's hardly an issue I think), but because more points mean shorter splines, and shorter splines are more likely to get disturbed when rounding of the control points take place. As it is now, both the new and the old algorithm produce pretty good approximations of the cubic splines, but once control points are rounded, it is not uncommon for disturbing bumps and corners to appear on the outline. (It may be that this is more probable to happen to autotraced glyphs.) My belief is that, since (as far as I know) converting to quadratic splines is done only with the intention of producing a TrueType font, ideally, a converting algorithm (be it based on the old or new one) should take the rounding of the control points into consideration when choosing what quadratic splines to produce... (I have tentatively started to try to develop one such algorithm, but I can't say if I'll be successful. ;-) ) Back to the new algorithm -- the attached font contains some glyphs, all of which make the latest Fontforge freeze when trying to convert them. Regards, Johan Winge |