Slow output in `tmux` in recent `xterm`

[agguser]

Issue description

When using tmux (v3.1c) in recent xterm (v365) (in Arch Linux), output of some programs is very slow (but fast if run in xterm outside tmux, or in tmux in urxvt or xterm v362).

$ xterm
$ perl -E 'say for 1..10000'   # fast output

$ tmux -f /dev/null
$ perl -E 'say for 1..10000'   # slow output
$ seq 1 10000   # fast output

Required information

$ uname -sp && tmux -V && echo $TERM
Linux unknown
tmux 3.1c
xterm-256color

[nicm]

Does this still happen with master?

[agguser]

Not happen with tmux next-3.3. Thanks.

Sorry, I forgot that I still use xterm v362.

The problem is still there with tmux next-3.3 in xterm v365.

[nicm]

So you change the xterm version and not the tmux version and xterm is now slow?

[nicm]

Show me the server log please.

[nicm]

Looks like xterm 365 is not responding to \033[>q, although it does so when I send it manually.

[nicm]

Oh no, it does respond. I am reading the wrong thing in the log.

[nicm]

xterm 365 works fine for me, show me the tmux server log please.

[agguser]

To be clear: tmux 3.1c and next-3.3 are slow in xterm 365, but fast in xterm 362. Tested with tmux -f /dev/null.

To produce the "tmux server log" (tmux-server-17401.log), I use tmux -f /dev/null -vv, and notice that the output is fast! I don't know how to produce the log without -v.

Let me repeat: in xterm 365, tmux -f /dev/null is slow, tmux -f /dev/null -vv is fast.

[nicm]

Logging doesn't change anything about how tmux writes to the terminal...

But perl is slow and seq is fast?

Can you do this:

$ perl -E 'say for 1..10000'  >perl.out
$ seq 1 10000 >seq.out

Then if you do cat perl.out is it still slow? Is cat seq.out still fast?

Please send me both files.

[agguser]

Both are fast (with or without -vv). And here are the server logs (with -vv): tmux-server-18106.perl.log, tmux-server-18116.seq.log.

By the way, I found that without -vv (where perl … is always slow), seq 1 10000 | sed 's/^//' is sometimes fast, sometimes slow. (You may try for i in {1..10}; do seq 1 10000 | sed 's/^//'; done.)

I don't have a clue why this happens, but all output are fast outside tmux or in xterm 362.

[nicm]

Run strace -tttt -fo perlstrace.out perl ... and the same with seq and show me the two strace files.

[nicm]

I think perl is slower and is generating output in smaller blocks so tmux does not have so much opportunity to throw it away, but there is no reason it should be hugely slower.

[nicm]

Did this definitely start in 365? Is 364 OK?