[PATCH] Assume "unknown" terminal type by default

[PATCH] Assume "unknown" terminal type by default

[Stephanos Ioannidis]

This patch modifies the `_rl_init_terminal_io` function to pass
`unknown` terminal type instead of `dumb` when no terminal name is
provided by the caller (e.g. when the `TERM` environment variable is
not set on Win32).

This ensures that the termcap provider (e.g. ncurses) resolves the
default preferred terminal type instead of using the `dumb` terminal
type.

Signed-off-by: Stephanos Ioannidis <root@stephanos.io>
---
 readline/readline/terminal.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/readline/readline/terminal.c b/readline/readline/terminal.c
index e557389776..086211d408 100644
--- a/readline/readline/terminal.c
+++ b/readline/readline/terminal.c
@@ -444,7 +444,7 @@ _rl_init_terminal_io (const char *terminal_name)
   tty = rl_instream ? fileno (rl_instream) : 0;
 
   if (term == 0)
-    term = "dumb";
+    term = "unknown";
 
 #ifdef __MSDOS__
   _rl_term_im = _rl_term_ei = _rl_term_ic = _rl_term_IC = (char *)NULL;
-- 
2.26.2

Re: [PATCH] Assume "unknown" terminal type by default

[Eli Zaretskii]

> From: Stephanos Ioannidis <root@stephanos.io>
> Date: Tue,  2 Jun 2020 12:52:49 +0900
> Cc: Stephanos Ioannidis <root@stephanos.io>
> 
> This patch modifies the `_rl_init_terminal_io` function to pass
> `unknown` terminal type instead of `dumb` when no terminal name is
> provided by the caller (e.g. when the `TERM` environment variable is
> not set on Win32).
> 
> This ensures that the termcap provider (e.g. ncurses) resolves the
> default preferred terminal type instead of using the `dumb` terminal
> type.

Isn't this something that ncurses itself should do on MS-Windows?

RE: [PATCH] Assume "unknown" terminal type by default

[Stephanos Ioannidis]

> Isn't this something that ncurses itself should do on MS-Windows?

ncurses assumes the "#win32con" terminal type (required for proper terminal 
operation on the Windows console) when the tgetent is called with 
termtype=NULL or termtype="unknown".

In this case, _rl_init_terminal_io is passing termtype="dumb" to tgetent when 
the TERM environment variable is not defined, so ncurses is resolving the 
"dumb" terminal type instead of required "#win32con".

This effectively causes some keys (e.g. backspace and arrow keys) to not work 
in the gdb command line, and this patch ensures that the problem is fixed.

p.s. I apologise for the horrible line formatting in the previous messages; I
 am sending these from a home machine and the mail client settings are all
 messed up.

Re: [PATCH] Assume "unknown" terminal type by default

[Eli Zaretskii]

[Please keep the mailing list on the CC.]

> From: Stephanos Ioannidis <root@stephanos.io>
> Date: Tue, 2 Jun 2020 16:03:27 +0000
> 
> > Isn't this something that ncurses itself should do on MS-Windows?
> 
> ncurses assumes the "#win32con" terminal type (required for proper terminal
> operation on the Windows console) when the tgetent is called with
> termtype=NULL or termtype="unknown".
> 
> In this case, _rl_init_terminal_io is passing termtype="dumb" to tgetent
> when the TERM environment variable is not defined, so ncurses is resolving
> the "dumb" terminal type instead of required "#win32con".
> 
> This effectively causes some keys (e.g. backspace and arrow keys) to not
> work in the gdb command line, and this patch ensures that the problem is
> fixed.

Ah, okay, I guess you are using a newer version of ncurses than I do.
(In my version ncurses only looks at $TERM.)

Thanks.

Re: [PATCH] Assume "unknown" terminal type by default

[Tom Tromey]

>>>>> "Stephanos" == Stephanos Ioannidis <root@stephanos.io> writes:

Stephanos> This patch modifies the `_rl_init_terminal_io` function to pass
Stephanos> `unknown` terminal type instead of `dumb` when no terminal name is
Stephanos> provided by the caller (e.g. when the `TERM` environment variable is
Stephanos> not set on Win32).

Stephanos> This ensures that the termcap provider (e.g. ncurses) resolves the
Stephanos> default preferred terminal type instead of using the `dumb` terminal
Stephanos> type.

Stephanos> Signed-off-by: Stephanos Ioannidis <root@stephanos.io>
Stephanos> ---
Stephanos>  readline/readline/terminal.c | 2 +-
Stephanos>  1 file changed, 1 insertion(+), 1 deletion(-)

Patches to readline should normally go to the readline maintainer first.
I'd prefer we not diverge from upstream unless it is truly necessary.
If he accepts the patch, we can check it in locally.

thanks,
Tom