Cursor skips a certain cell visually under certain conditions

[megahomyak]

I've been writing my TUI app and noticed that a certain cell (bottom row, fifth from right) gets skipped under certain very specific conditions that I've been trying to pin down, but failed (the bug is not present in one case in reproduction, and under the same circumstances it is present in my TUI app). After the skip, the cursor still keeps having its offset incremented by "1", even if it's move()d after the skip's beginning, and goes off-screen if the last character of the line is selected. It still functions as if it's correctly placed, though

Repro: https://github.com/megahomyak/cursesfuck/tree/54cbc6fde2f55883fa6102b8c8631b0daeadc50f/kitty_bug

The TUI app: https://github.com/megahomyak/megatui/tree/655f680549785e216745777b9df536adcc85e43f

The bug is NOT present in the repro or in the app in: Alacritty, Konsole, Terminator, Termux

Screenshots:

The TUI app:

The repro:

[kovidgoyal]

This is with what version of kitty on what OS and did you try it with
kitty --config NONE?

[megahomyak]

I am very sorry, I forgot to provide that information

General parameters

• OS: Ubuntu 24.04.2 LTS over Linux 6.8.0-63-generic
• Architecture: x86_64

When I was making the issue

• Kitty: 0.32.2 (downloaded through apt, it's still "latest" there, "0.32.2-1ubuntu0.4")
• --config NONE: the issue persists

After updating Kitty

• Kitty: 0.42.2 (from Github releases)
• --config NONE: the issue persists

[kovidgoyal]

This is a bug in your program, easily demonstrated by running kitty with
kitty --dump-commands | grep screen_cursor_position

(note that --dump-commands buffers output so you will need to quit kitty
to see the full log or you can patch it with the following)

diff --git a/kitty/boss.py b/kitty/boss.py
index 8f898f80b..e686edb8e 100644
--- a/kitty/boss.py
+++ b/kitty/boss.py
@@ -282,7 +282,7 @@ def fmt(x: Any) -> Any:
                 if isinstance(x, dict):
                     return json.dumps(x)
                 return x
-            safe_print(what, *map(fmt, a))
+            safe_print(what, *map(fmt, a), flush=True)
 # }}}

Your program (or rather ncurses since you are using that)
is sending an incorrect cursor position escape code at the magic
position.

As general advice, I would suggest you avoid ncurses like the plague. It
is a truly bad library. You can write modern terminal programs without
using any libraries but if you must use a library, please use a better
one. There are many, such as notcurses, for example.

[kovidgoyal]

Actually, no never mind I looked at the commands more closely and ncurses is actually moving by moving the cursor to the right edge of the screen, drawing a space in no wrap mode and then backspacing. Utterly insane, but I can possibly get kitty to handle this as ncurses expects.

[megahomyak]

Your program (or rather ncurses since you are using that)
is sending an incorrect cursor position escape code at the magic
position.

Got it. I'm sorry I thought it was Kitty

As general advice, I would suggest you avoid ncurses like the plague. It
is a truly bad library. You can write modern terminal programs without
using any libraries but if you must use a library, please use a better
one. There are many, such as notcurses, for example.

Thank you so much!! Noted. I will try to use other tools

Utterly insane, but I can possibly get kitty to handle this as ncurses
expects.

Thank you very much for such readiness to fix issues, but just in case, if
other approaches, including raw escape codes, wouldn't have this issue, then I
don't think it's much of a priority - I will try to rewrite that little piece of
software using something less quirky

[kovidgoyal]

Well I have made the fix, let's see how it goes. It has some crazy side-effects like if you input in cooked mode, for example by running the read command in a shell, and type all the way to the right edge of the screen and then press backspace, what you see on the screen is no longer what's present in the kernel buffer. This happens on all terminals that have this behavior. Absurd. This should really be fixed in ncurses, there is no reason for it to use backspace to position the cursor from the right edge of the screen. However, since most terminals actually behave this way, in the interests of compatibility, I will leave the fix in for the time being and see.

[kovidgoyal]

Actually, I found a better fix. There is a terminfo property that gets ncurses to not do this and that property means that backspace now works to move to the previous line in cooked mode, which is a nice bonus.