[MPEG-OTSPEC] Some research on TT instructions and cubics

[Laurence Penney]

Dear Hin-Tak,

Please review the official proposal document (not the .md file I referred to previously), in particular page 3, where the language used may be more to your liking ("must" means no need to speak of "undefined"):

https://github.com/harfbuzz/boring-expansion-spec/blob/main/iso_docs/WG03_otf-improvements.pdf

And please start a new thread.

- Laurence

> On 23 Sep 2023, at 12:12, Hin-Tak Leung <htl10 at users.sourceforge.net> wrote:
> 
> Also, not having the implicit on-curve point nullifies claims of size-benefits of cubic over quadratic - if you use quadratic * properly * with implied on-curve points, it is size-efficient.
> 
> Poorly-done quadratics is poorly-done. Better than that, means nothing.
> 
> On Saturday, 23 September 2023 at 11:53:42 BST, Hin-Tak Leung <htl10 at users.sourceforge.net> wrote: 
> 
> 
> That's quite unacceptable, to be honest, with a lot of "undefined behaviour". What will happen, if it takes off, is that quirks and limitations of one particular implementation will come to be expected, then get formalised as "limitations/quirks as specification" :-(.
> 
> Just telling people not to have not-2 consecutive off-curve points, doesn't mean they won't try. Then, they expect it to behave like their favourite implementation does, and try to push it into the spec... I think I can say there is such a past instance where maths were done in 16-bit due to tech limitations, then people start to expect, and * demand * that maths must overflow in specific ways.
> 
> You can say it is restricted now, but pretty sure that some will not only ignore the restriction, but also * demand * that the restriction is to be broken in quite specific ways, to  their preference / convenience.
> 
> It is also a hugh step backwards - in quadratics, contours consisting of entirely off-curve points are legal, and in fact beneficial for a few reasons.