Extremely slow output causing unresponsive tab

The problem is the same as described in #6394 which should have been fixed.

I encountered the same problem with .179 on Windows 11 x64, sometimes the symptom is much worse —

Tabby enters a status not responding to user inputs for the exact tab (while still printing lots of lines quickly overflow the scroll buffer); AND Tabby starts to use a significant amount of CPU (on a very high-end PC); AND Even if I repeatedly enter something like Ctrl-C immediately after entering the command that triggers a lot of output, it takes 10 seconds or 20 seconds or even longer to stop the output (especially when the output lines are colored). I tried rolling back to .176 and it seems better (still in verification).

A typical command to trigger the issue fgrep . -Inr / --color.

I can grab and attach a performance trace later (after I have time to reinstall .179 debug it) if it sounds necessary to you.

[wudisheng]

Hi,

I’ve got some time to collect the performance profiles.

The way I collect these profiles is as below

• Download portable versions of 1.0.176 and 1.0.179 (while 1.0.176 is still installed system-wide);
• Prepare a directory code containing about 200MB source files;
• For each of the portable versions above, and in a default WSL shell
  • Run command time ( find code | grep --color a ) several times (warm-up, to eliminate potential filesystem caching effects);
  • Start recording performance trace;
  • Run command time ( find code | grep --color a ) 3 times in a row, each time immediately after seeing the command prompt shows up (i.e. the output finishes scrolling and rendering);
  • Stop recording performance trace and save the profiling results;
• Compress both profiles for comparison and upload.

The difference is significant — the command itself executes for about 0.5 second (as reported by time), and

• For version 1.0.176, it took about 1 second to finish rendering;
• For version 1.0.179, it took about 19 seconds, and maybe using more CPU (since the noise from adaptive fans is obviously louder than .176);

No configuration changed during the benchmark.

Let me know if you need more performance traces with certain actions (e.g. pressing Ctrl-C immediately after entering the command in .179).

Profile-1.0.176.json.gz Profile-1.0.179.json.gz

[wudisheng]

Thanks a lot!