Kitty: Wrong render of some unicode symbols

[webhive]

I am using rounded corners symbols in my terminal. It work fine with Gnome Terminal, but kitty render these symbols (U+256D, U+2570) wrong way - for some reasons it is not rounded

Here is a screenshoot

Initially thought it was a font issue - but it is not depends on font. And Gnome Terminal used exactly the same font and everything displayed well.

[kovidgoyal]

It's not a font issue. Box drawing characters are not drawn with fonts otherwise they dont always line up correctly. I simply haven't bothered with implementing rounded corners, in the custom drawing code. Patches are welcome, relevant code is in box_drawing.py search for TODO

[webhive]

Done some quick attempt using midpoint algorithm
It is harder work when initially expected :) Definitely need to implement somehow a thinkness and antialiasing.

[Luflosi]

https://en.wikipedia.org/wiki/Xiaolin_Wu%27s_line_algorithm may help. Also see 3fa7007e910242c4fb3f691acae74b887c26a718.

[kovidgoyal]

Yeah that's why I have not bothered implementing it. There are examples
of doing curves and anti-aliasing in that file. For instance, the D()
function or fill_region()

[webhive]

Ha! Added thikness.

Small font


Large font

So the only thing to do is antialiasing. Damn - I never had so fun for a long time :)

[webhive]

Hmm - applied Wu AA

Top right is mine, another one is GnomeTermnal

Close look

It is definitely antialiased, but looks ugly. Gnome Terminal has better AA while it is way more simple. Any suggestions?

[kovidgoyal]

It's been a while since I did any anti-aliasing, but as I recall using
the various theoretical methods was overkill/didn't work well. Just a
simple y-axis blur seemed to work best, search the file for anit-alias
to see the implementation I used.

[Luflosi]

@webhive maybe you didn't implement the algorithm correctly? https://create.stephan-brumme.com/antialiased-circle/ has some sample code for using Wu's algorithm to draw a circle. But I'm not sure how to make the line thicker than one pixel. Maybe the approach @kovidgoyal mentioned is better. Or maybe you find a completely different algorithm.

[webhive]

@Luflosi Funny - used exactly this article. Yea - will try @kovidgoyal approach