Re: Bugs when porting ncurses to a new platform

On Wednesday, July 1st, 2026 at 2:47 AM, Thomas Dickey <dickey@invisible-is=
land.net> wrote:

> On Tue, Jun 30, 2026 at 12:27:45AM +0000, Antonio Ni=C3=B1o D=C3=ADaz wro=
te:
> > Hello!
> >=20
> > I'm trying to get ncurses working in a new platform (Nintendo DS using
> > [BlocksDS](https://blocksds.skylyrac.net/)).  I've created a terminal
> > emulator that supports 16-color, 256-color and direct color commands
> > (foreground and background), as well as a few other commands like "bold=
",
> > "move cursor" and "clear screen".  Basically, I'm cherry-picking termin=
fo
> > configuration settings from other terminals to create my own configurat=
ion
> > based on what I actually support, and I'm trying to understand every se=
tting
> > I use.  So far I've managed to get 16 and 256 color modes working, as w=
ell as
> > moving the cursor around and clearing the screen.  However, I have two =
issues
> > that I've spent a few days debugging with no success.  I've even added =
printf
> > logs in ncurses in many places (the trace system doesn't work for some
> > reason) but ncurses has so many build options and code paths that it's =
hard
> > to know what actually gets used (printf-debugging has helped to get col=
or
> > working, though).
> >=20
> > ----------------
> ...
> > 2) getch() only works for blocking input.
> >=20
> > If I use `timeout(-1)` I can see calls to `read()`.  They block as expe=
cted
> > and I can see that the value is received by ncurses.  However, if I use
> > `timeout(0)` I never see any syscall being used.  Maybe I'm looking at =
the
> > wrong place, but the same code works on PC, so I assume it's something =
to do
> > with my port.  I do support setting non-blocking mode for `stdin` with
> > `ioctl(FIONBIO)` and `fcntl(F_SETFL, O_NONBLOCK)`, and I also support
> > querying the amount of characters available to read from `stdin` with
> > `ioctl(FIONREAD)`, but I don't know if this is even used by ncurses.
>=20
> no - ncurses basically uses select() or poll() to detect new data.
>=20
> In a quick check, timeout() sets a _delay member of the window,
> which when negative will bypass a timed delay when looking for data.
> =20
> > How does ncurses get new characters for non-blocking input?
> >=20
> > ----------------
> >=20
> > This is my build configuration:
> >=20
> > ./configure \
> > --with-normal \
> > --without-shared --without-progs --without-manpages --without-tests \
> > --target=3Darm --host=3Darm --build=3Damd64 \
> > --disable-database --disable-db-install \
> > --disable-gnat-projects --disable-home-terminfo \
> > --enable-termcap \
> > --disable-widec \
> > --prefix=3D$PWD/mybuild \
> > --with-trace \
> > --enable-getcap \
> > --disable-sigwinch \
> > --enable-ext-colors --enable-rgb-color \
> > --enable-ext-mouse \
> > CFLAGS=3D"-mthumb -mcpu=3Darm946e-s+nofp -O2 -ffunction-sections -fdata=
-sections -I$BLOCKSDS/libs/libnds/include" \
> > BUILD_CC=3Dgcc \
> > CC=3Darm-none-eabi-gcc \
> > cf_cv_have_tcgetattr=3Dyes \
> > ac_cv_func_gettimeofday=3Dyes \
> > ac_cv_type_sigset_t=3Dno
>=20
> The output from the configure script (config.log, config.status) would
> tell us for example if it's using select or poll.

Ah, yes, that seems to be it, my select() and poll() don't work with stdin,=
 so
it can't detect the available characters.

Unfortunately, implementing this isn't exactly trivial (I support select() =
and
poll() for network sockets with lwIP, so I need a way to redirect select() =
calls
to lwIP and to my own stdin handler somehow) so I might just add a small pa=
tch
to add an extra case to check_pending() to use ioctl(FIONREAD) if neither o=
f the
functions are available. There's already a custom case for BeOS (which also
supports them also for networking), but it's not exactly the ioctl() I need=
.

Thanks a lot, I'll investigate how I can work around this limitation!

Antonio