Re: COLOR_PAIRS on Fedora 29

[Timothy Allen]

On Wed, 2019-01-16 at 10:20 -0600, Bryan Christ wrote:
> I'm debugging an issue on Fedora 29.  The terminfo value for pairs on
> xterm-256color is set to 0x10000 which is in the range of an unsigned
> short but not a short.  Isn't this problematic since most the ncurses
> APIs like wattr_set() expect a short?

In the ncurses 5 ABI, the terminfo database file-format uses signed
shorts for everything, so values are clamped at 0x7FFF even if
logically they should be higher.

In the ncurses 6 ABI, the terminfo database file-format has been
extended to allow larger numbers, so values generally do not need to be
clamped.

For example, the new "xterm-direct" terminal description (xterm with
24-bit colour) reports colors=0x1000000, although it still has
pairs=0x10000 because the actual number of possible pairs is a 49-bit
number, and ncurses *can't* allocate that many pairs just for memory-
allocation reasons.

As for wattr_set(), the manpage says:

> Each of the functions added by XSI Curses has a parameter opts, 
> which X/Open Curses still  (after  more  than twenty  years) 
> documents as reserved for future use, saying that it should be 
> NULL.  This implementation uses that parameter in ABI 6 for the 
> functions which have a color-pair parameter to support extended 
> color pairs:
>
> •   For functions which modify the color, e.g., wattr_set, if opts 
>     is set it is treated as a pointer to  int, and used to set the 
>     color pair instead of the short pair parameter.

> I could simply change my code to an unsigned short, but i'm not sure
> what that would do to ncurses.  Alternatively, I could simply
> redefine COLOR_PAIRS to 0x7FFF which is what Bionic Beaver uses.  As
> best I can tell, it looks like Fedora 29 made a poor decision.

The current Debian Testing uses the ncurses 6 ABI, as does Ubuntu
Cosmic, so it's not just Fedora 29.