On Sun, Nov 23, 2025 at 05:35:02PM +0100, Sven Joachim wrote: > On 2025-11-23 16:48 +0100, Chris Hofstaedtler wrote: > > > Control: retitle -1 ncurses: API/ABI break for tigetflag return value on > > unsigned char archs > > Control: severity -1 serious > > > > * Chris Hofstaedtler <[email protected]> [251123 16:45]: > >>I see patch 20251101 changed the definition of ABSENT_BOOLEAN from > >> ((signed char)-1) to > >> ((NCURSES_SBOOL)-1)
...fixing a compiler warning :-( > >> While configure.in detects the possibility of using `signed char` > >> for `NCURSES_SBOOL`, I think by default it uses `char` (without > >> --enable-signed-char). as I recall it, the signed char type wasn't well standardized ~ 20 years ago, but that Debian needed it. So it was optional. > > I think this is an unintentional ABI break on unsigned-char archs, and > > also all binaries which were compiled using 20251101 or newer need to > > be recompiled(!). > > I do not think it is that bad, because ABSENT_BOOLEAN is defined in > tic.h which nobody uses. There will likely be more runtime problems > because of the incorrect tigetflag() return value for missing > capabilities, though. Actually it's used in lib_ti.c, and promoted to an int. So it may/may not be sign-extended, which I assume is the bug. I have an arm64 machine, which may show me the issue. -- Thomas E. Dickey <[email protected]> https://invisible-island.net
signature.asc
Description: PGP signature
