Kitten `clipboard` non-deterministically copies/truncates big scrollback when piped tab `@text` by keybinding

[ronjouch]

To copy the whole scrollback to the clipboard, I use

map kitty_mod+a pipe @text tab kitty +kitten clipboard

With a small scrollback it works great, but with a big scrollback it consistently fails to copy the whole scrollback, instead truncating the end of the scrollback at non-deterministic positions 😕.

For example, after ls -la /usr/bin (2800 lines), when pressing kitty_mod+a,

1. First press yields 2122 lines
2. Second press yields 317 lines
3. Third press yields 2635 lines

I'm using kitty master @ 4faf71b under Ubuntu 19.04, my shell is fish 3.0.2 .

---

Note: this is a followup of [Help] Map key to copying whole scrollback to clipboard, where I asked for advice to achieve the same result, using xclip/xsel. It worked, until it didn't, for reasons I didn't investigate, so I moved to kitten clipboard and was happy, then I noticed this problem.

[kovidgoyal]

Does using the clipboard kitten directly with a large text file work?

kitty +kitten clipboard < some text file

[ronjouch]

Does using the clipboard kitten directly with a large text file work?

kitty +kitten clipboard < some text file

@kovidgoyal yes this works consistently (same test case, with a ls.txt containing the same 2800 lines of ls -la /usr/bin/). Problem happens only when invoking the kitten from a binding, which opens a transient tab which gets closed after a few milliseconds. Maybe the transient tab is closed too soon / fails to wait for the kitten to finish?

And with that additional information, maybe my initial xsel-based solution actually suffers from the same problem?

As usual, thanks for the fast feedback.

[ronjouch]

Additional data point: this feels like a regression, I don't think I did encounter that back a few months ago. @kovidgoyal I could try a bisect, if that's helpful.

[kovidgoyal]

Should be easy to fix, create a script that calls the kitten and sleeps
for a little while afterwards then use that in your binding.

[ronjouch]

@kovidgoyal I compiled 84fcabe and updated my config to map kitty_mod+a pipe @text tab kitty +kitten clipboard --wait-for-completion. Although it visibly waits, it now fails differently, truncating the beginning of a big scrollback and only keeping the end. In a terminal with a big scrollback of several ls -la /usr/bin, I get three only-slightly different outputs in my clipboard:

1. First run: 11871 lines, 744.285 bytes.
2. Second run: 11878 lines, 744.758 bytes.
3. Third run: 11876 lines, 744.638 bytes.

My scrollback_lines is currently set to 100000, so it shouldn't be the problem.

Am I missing something?

[kovidgoyal]

Just use

pipe @text clipboard dummy

[ronjouch]

Just use

pipe @text clipboard dummy

@kovidgoyal yes! That works muuuuuch better, it's reliable, flicker-less, and fast. Sorry, had focused on the first commit with --wait-for-completion but didn't see the contents of the second, adding the pipe command to send to clipboard.

Thanks for the fix and for the continued work on kitty 🙏 ! Good day/night.