I've actually heard a plausible theory about the TUI being janky, that being that they avoid use of the alternate screen feature of ANSI (and onwards) terminals.

The theory states that Anthropic avoids using the alternate screen (which gives consuming applications access to a clear buffer with no shell prompt that they can do what they want with and drop at their leisure) because the alternate screen has no scrollback buffer.

So for example, terminal-based editors -- neovim, emacs, nano -- all use the alternate screen because not fighting for ownership of the screen with the shell is a clear benefit over having scrollback.

The calculus is different when you have an LLM that you have a conversational history with, and while you can't bolt scrollback onto the alternate screen (easily), you can kinda bolt an alternate screen-like behaviour onto a regular terminal screen.

I don't personally use LLMs if I can avoid it, so I don't know how janky this thing is, really, but having had to recently deal with ANSI terminal alternate screen bullshit, I think this explanation's plausible.

▲

edvinbesic 3 hours ago | parent | next [-]

Not disagreeing but scrolling works just fine in vim/emacs/etc. Wouldn't it be just managing the scroll back buffer yourself rather than the terminals?

▲

jdiff 3 hours ago | parent | next [-]

Yes, but this does come with differences and tradeoffs. If the terminal isn't managing the scrollback, you don't get scrollbars and you lose any smooth/high resolution scrolling. You also lose fancy terminal features like searching the scrollback, all that needs to be implemented in your application. Depending on the environment it can also wind up being quite unpleasant to use with a trackpad, sometimes skipping around wildly for small movements.

▲

_verandaguy 3 hours ago | parent [-]

The other part (which IMO is more consequential) is that once the LLM application quits or otherwise drops out of the alternate screen, that conversation is lost forever.

With the usual terminal mode, that history can outlive the Claude application, and considering many people keep their terminals running for days or sometimes even weeks at a time, that means having the convo in your scrollback buffer for a while.

	▲	jaredsohn 2 hours ago \| parent [-]
		>that conversation is lost forever. You should be able to find it in ~/.claude You can also ask Claude to search your history to answer questions about it.

▲

bombela 3 hours ago | parent | prev [-]

I think they were saying that in "cup" screen mode (CUP: CUrsor Position, activated with smcup termcap), when you exit (rmcup) the text is lost, as well as the history since it was managed by the application, not the terminal.

Their hypothesis was that maybe there was aj intention to have claude code fill the terminal history. And using potentially harzardous cursor manipulation.

In other words, readline vs ncurse.

I don't see python and ipython readline struggling as bad tho...

▲

jlokier 2 hours ago | parent | prev | next [-]

I don't think that's likely to explain jankiness. I do know my way around terminal screens and escape codes, and doing flicker-free, curses-like screen updates works equally well on the regular screen as on the alternate screen, on every terminal I've used.

It's also not a hard problem, and updates are not slow to compute. Text editors have been calculating efficient, incremental terminal updates since 1981 (Gosling Emacs), and they had to optimise better for much slower-drawing terminals, with vastly slower computers for the calculation.

▲

dantillberg 3 hours ago | parent | prev [-]

Yesterday, I resumed a former claude code session in order to copy code it had generated earlier in that session. Unfortunately, when resuming, it only prints the last N hundred lines of the session to the terminal, so what I was looking for was cut off.

I think that for this sort of _interactive_ application, there's no avoiding the need to manage scroll/history.

	▲	pjeide 2 hours ago \| parent [-]
		That conversation should still exist in the Claude Code log files. Just give Claude some context on how to find it, and it will pull whatever you need. I use this to recall particularly effective prompts later on for reuse.