Printing to the scrollback buffer (or creating a managed scrollback buffer)

[welkinbai]

My question

When the text output by my UI exceeds one screen, scrolling up the terminal causes the previous text to disappear.
I am using this framework to build a terminal-based AI application, and I hope to retain the previous chat history when the user scrolls the screen. What should I do?

[zyriab]

Use tea.Print[ln|f] and print directly to stdout.
You can use the regular rendering (View()) for the stuff at the bottom.

[welkinbai]

Use tea.Print[ln|f] and print directly to stdout. You can use the regular rendering (View()) for the stuff at the bottom.

@zyriab Thank you for your suggestion, but I encountered a problem when trying it out. The responses from my AI assistant are streamed, but I found that whether I use tea.Println or tea.Printf, my streamed output gets appended with a newline character.
For example, if I have a reply: Hello, how can I help you?
The actual call might be:
tea.Printf("%s", "Hello, ")
tea.Printf("%s", "how can")
tea.Printf("%s", " I help")
tea.Printf("%s", " you?")
I hope there are no line breaks, but what I actually see is:
Hello,
how can
I help
you?

I feel that the bubbletea code doesn't seem to strictly follow its semantics when implementing tea.Printf. When the printLineMessage arrives, it is always treated as a new line and placed into queuedMessageLines. Am I right?

[zyriab]

You have to append each received token into an area that is rendered through View then once you get the next response, you flush the contents of that area to stdout.

It's tricky and there are a lot of edge cases but it's doable ;)

As for tea.Printf, yeah, it's actually printing into its own line every time. I think that's just a limitation of stdout. It's actually documented iirc

EDIT: I'm actually working on a similar project. I've done what I describe above but have met some issues (markdown rendering with Glamour thinks tokens are whole documents, token stream area height being limited by the viewport's, etc).
If I had to redo it from scratch, I would probably render everything in an alt-screen, viewport or something and mimic stdout.
Then, on app exit, I would take the whole content of the fake stdout and print it to stdout so that it stays explorable after.

I'm learning Bubble Tea as I go so take my advice with a pinch of salt! 😅

EDIT 2: I've tried rewriting my implementation with a Viewport but it doesn't allow text selection (or you have to forego scrolling). It's tricky, if I had time I would probably base myself on Crush PR and add support for it.

I'm also having issues with the token streaming area adding some artifacts from the rendered TUI parts on the stdout when I flush it. A workaround that might be to do like Claude Code and actually print full sentences/blocks to stdout without streaming each token in.
That's what I would do if the streaming is not super important in your desired UI.

Good luck!

[welkinbai]

@zyriab I agree with your thoughts, and you've gone further than I have, already trying several solutions I wanted to pursue, which is great 👍. Maybe you're right, and I should give up on streaming output first. Maybe I won't be able to resist the temptation and still want to try to solve it 😂.

[zyriab]

You should, it's a good challenge!
I've got it pretty much nailed down (using the streaming area that flushes to stdout).

Just having issues with artifacts from the rendered parts of the app (text input, mostly) ending up on stdout but I figured out that it came from the terminal emulator missing a refresh from Bubble Tea while scrolling very fast (when restoring conversations).
The workaround might be to either hide the text input when restoring or just restoring everything as a one time print, like Claude Code does.