Re: How should applications use direct-color mode?

[Thomas Dickey]

On Sun, Apr 15, 2018 at 03:05:06PM +1000, Tim Allen wrote:
> Kakoune[1] is an ncurses-based text editor that supports
> syntax-highlighting themes, and therefore wants to display a wide variety
> of colours in the terminal. Currently[2], for terminals where COLORS > 16
> and can_change_color() is true, it will allocate the required colours with
> init_color(), and otherwise it will try to pick the closest match from the
> first COLORS entries in a copy of the xterm default 256-colour palette.
> 
> Now that ncurses 6.1 has been released with support for direct-colour
> terminals, I'm investigating what would be required to teach Kakoune
> (or any application, really) how to use it.
> 
> As I understand it, the first change is that the application needs to
> be updated to call init_extended_pair() instead of init_pair(). This
> function is only present when NCURSES_EXT_FUNCS is defined, so the
> application can use #ifdef to choose which function is called.

#ifdef and #define are suitable tools...
 
> The second change is that instead of passing palette entries to
> init_extended_pair(), the application should pass actual packed RGB
> colours, such as 0xFF7F00 for orange. However, curs_color(3X) says

You could interpret it that way.  The existing standards as such don't
get that specific.

> "color values are expected to be in the range 0 to COLORS-1, inclusive
> (including 0 and COLORS-1)", and at least with the ncurses 6.1+20180210-1
> library from Debian Experimental "TERM=xterm-direct tput colors" prints
> 32767, or 0x7FFF. This implies that direct-color mode can only display

That's because it's an application using the 16-bit values.
A 32-bit terminfo application would probably give a different result.

> shades of blue, teal and green, and indeed that's the behaviour of
> ncurses demos like "dots.c". Is it intended that COLORS=32767 be
> interpreted as a sentinel value meaning "unlimited" rather than being
> a specific limit?

no - 32767 happens to be the largest value in a signed 16-bit number.
 
> The big question of course is: how can an application determine whether
> the terminal suppors direct colour or palettized colour? I can't seem
> to find any specific advice in curs_color(3x), and poking through the
> ncurses source-code suggests that the relevent terminfo field ("RGB") is
> used to avoid setting up a colour table (in start_color()) and in the
> implementation of extended_color_content() but nowhere else.

You could have "direct-color" with one bit per r/g/b.
It's merely a way of organizing the data.
 
> Thus I assume that an application that wants to support direct-color
> with ncurses 6.1 should do something like this:

...something like that (the developers will have to deal with the issues,
given the existing demos).
 
> Is that reasonable? Am I missing anything?
> 
> [1]: http://kakoune.org/
> [2]: https://github.com/mawww/kakoune/blob/master/src/ncurses_ui.cc#L151

By the way, this program is referencing ncurses internals:

https://github.com/mawww/kakoune/blob/master/src/ncurses_ui.cc#L12

-- 
Thomas E. Dickey <address@hidden>
https://invisible-island.net
ftp://ftp.invisible-island.net

signature.asc
Description: Digital signature