Arch Linux

Description: The "dialog" package treats BACKSPACE key as DELETE in the middle of a word.

Additional info:
* package version: ncurses 6.3-2, dialog 1:1.3_20220117-1

Steps to reproduce:
1. Run "dialog --inputbox Input 10 30"
2. Enter the word "abc" into the inputbox
3. Hit the LEFT-ARROW key twice to move the cursor over "b"
4. Hitting the BACKSPACE key deletes "b" instead of "a", essentially functioning like the DELETE key.

Workaround:
$ infocmp | sed -e 's/kbs=^H/kbs=^?/' >foo; tic foo

Analysis by Dialog's author, Thomas E. Dickey:

== BEGIN QUOTE ==

When I implemented the current scheme of key-bindings in 2005,
it seems that I preserved several pre-existing hard-coded definitions,
where one would expect just the definition in curses. But by doing
that, it's possible to have conflicting bindings.

Here's a chunk from doing that (i.e., replacing "8" from the source):

- case 8:
- case KEY_BACKSPACE:
- if (offset) {
- offset--;
- len--;
- for (i = offset; i <= len; i++)
- input[i] = input[i + 1];
- }
- break;

What I have now is

calendar.c:693: DLG_KEYS_DATA( DLGK_GRID_LEFT, CHR_BACKSPACE ),
dialog.h:164:#define CHR_BACKSPACE DLG_CTRL('H')
dlg_keys.h:122: DLG_KEYS_DATA( DLGK_DELETE_LEFT, CHR_BACKSPACE ), \
rangebox.c:175: DLG_KEYS_DATA( DLGK_FIELD_PREV, CHR_BACKSPACE ),
timebox.c:157: DLG_KEYS_DATA( DLGK_FIELD_PREV, CHR_BACKSPACE ),

and that dlg_keys.h definition is what you're using:

/* ^U == 21 */
#define INPUTSTR_BINDINGS \
DLG_KEYS_DATA( DLGK_BEGIN, KEY_HOME ), \
DLG_KEYS_DATA( DLGK_DELETE_ALL, CHR_KILL ), \
DLG_KEYS_DATA( DLGK_DELETE_LEFT, CHR_BACKSPACE ), \
DLG_KEYS_DATA( DLGK_DELETE_LEFT, KEY_BACKSPACE ), \
DLG_KEYS_DATA( DLGK_DELETE_RIGHT, CHR_DELETE ), \
DLG_KEYS_DATA( DLGK_DELETE_RIGHT, KEY_DC ), \
DLG_KEYS_DATA( DLGK_FINAL, KEY_END ), \
DLG_KEYS_DATA( DLGK_GRID_LEFT, KEY_LEFT ), \
DLG_KEYS_DATA( DLGK_GRID_RIGHT, KEY_RIGHT )

However, the KEY_xxx symbols (from curses) override the CHR_xxx symbols.

Using infocmp, you can see what kbs (KEY_BACKSPACE) is set to.

I can see by that, that Arch has kbs=^H for xterm, which is incorrect
(for Linux...).

But CHR_DELETE is ^?, which is not mapped in curses.

DLG_KEYS_DATA( DLGK_DELETE_RIGHT, CHR_DELETE ), \

The value used for kbs is a configure option because Linux differs from
almost everyone else (Cygwin does imitate Linux):

--with-xterm-kbs=XXX
Configure xterm's terminfo entries to use either BS (^H, i.e., ASCII
backspace) or DEL (^?, or 127). XXX can be BS (or bs, 8) or DEL
(or del, 127).

During installation, the makefile and scripts modifies the "xterm+kbs"
terminfo entry to use this setting.

xterm attempts to use the system's convention for the "backspace" key.
It looks like Arch didn't use the configure option:

https://github.com/archlinux/svntogit-packages/blob/packages/ncurses/trunk/PKGBUILD

fwiw, Debian gets that right:

https://salsa.debian.org/debian/ncurses/-/blob/master/debian/rules

I'd have done the same for the myriad of xterm-lookalikes, but most of
those are hard-coded to imitate xterm running on Linux (and they generally
are misconfigured in their ports to the BSDs and Unix platforms).

> May I know if this is expected behaviour? Is there any way I can make it do
> an actual BACKSPACE instead of DELETE?

As a workaround (a bug report to Arch might help), you can do something like

infocmp | sed -e 's/kbs=^H/kbs=^?/' >foo; tic foo

== END QUOTE ==