[BUG] weird encoding issues on (ghostty) terminals

Describe the bug

When running local uc commands I get weird artifacts on my screen

% ./uncloud machine ls
^[[?2026;2$y^[[?2027;1$yNAME        STATE   ADDRESS         PUBLIC IP   WIREGUARD ENDPOINTS                      MACHINE ID
machine-1   Up      10.210.0.1/24   -           172.19.0.2:51820, 131.174.139.88:51820   656c506c2da5e6012a3f20a157d709fe

Expected behavior

Don't print any escapes

Environment:

asahalinux fedora32/ghostty 1.3.1

Uncloud versions: main from github

I've seen the color and palette stuff in other PRs, my hunch is, this is not compatible. An option to disable all these colorization would also be nice (i'm colorblind, so I also want different colors)

This seems to be a "This type of sequence is commonly referred to as a CSI (Control Sequence Introducer)"... wat

tried different terminals, term infos, shells and shell prompts, but doesn't seem to be that.

% ./uncloud machine ls | cat fixes it. It's the connectClusterWithProgress which uses tea which breaks on fedora for some reason

it's this charmbracelet/bubbletea#1590

Bubbletea/lipgloss should support NO_COLOR=1 (https://no-color.org/) to disable colours in the output out of the box.

But this is only about colours, other control characters could still be used, e.g. for printing an interactive progress when connecting to the cluster.

It would be helpful if you can try some of the fixes linked in charmbracelet/bubbletea#1590.

I was not able to reproduce this in iterm2/ghostty/Terminal using fish/bash/zsh on macOS.

i think I just have to sit it out and wait for it to be fixed upstream

diff --git i/go.mod w/go.mod
index b278c90..c59f08a 100644
--- i/go.mod
+++ w/go.mod
@@ -2,6 +2,8 @@ module github.com/psviderski/uncloud

 go 1.26

+replace charm.land/bubbletea/v2 => github.com/saltydk/bubbletea/v2 v2.0.0-20260318220615-96c1e05cb8c1
+
 require (
        charm.land/bubbles/v2 v2.0.0
        charm.land/bubbletea/v2 v2.0.2
diff --git i/go.sum w/go.sum
index 2a1ef70..de0fb70 100644
--- i/go.sum
+++ w/go.sum
@@ -1078,6 +1078,8 @@ github.com/russross/blackfriday/v2 v2.0.1/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQD
 github.com/russross/blackfriday/v2 v2.1.0 h1:JIOH55/0cWyOuilr9/qlrm0BSXldqnqwMsf35Ld67mk=
 github.com/russross/blackfriday/v2 v2.1.0/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQDYRxCVz55jmeOWTM=
 github.com/ryanuber/columnize v0.0.0-20160712163229-9b3edd62028f/go.mod h1:sm1tb6uqfes/u+d4ooFouqFdy9/2g9QGwK3SQygK0Ts=
+github.com/saltydk/bubbletea/v2 v2.0.0-20260318220615-96c1e05cb8c1 h1:PQLktKXSoxCBczHOJWjAIi0dDlSx2zZ6/b5kzIOSmFs=
+github.com/saltydk/bubbletea/v2 v2.0.0-20260318220615-96c1e05cb8c1/go.mod h1:ukj5SpUwpemB+rLyqsw+R6SIDaQE53fMxZsojTUQrxc=
 github.com/santhosh-tekuri/jsonschema/v6 v6.0.1 h1:PKK9DyHxif4LZo+uQSgXNqs0jj5+xZwwfKHgph2lxBw=
 github.com/santhosh-tekuri/jsonschema/v6 v6.0.1/go.mod h1:JXeL+ps8p7/KNMjDQk3TCwPpBy0wYklyWTfbkIzdIFU=
 github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0=

fixes it, but that is not the official fix (and not the official repo!)

% ./uncloud machine list
NAME        STATE   ADDRESS         PUBLIC IP   WIREGUARD ENDPOINTS                      MACHINE ID
machine-1   Up      10.210.0.1/24   -           172.19.0.2:51820, 131.174.139.88:51820   58de29fd9f276a1cc05d95504e2ff5aa

with the control master stuff, this is now happening all the time and it's driving me nuts. I'll ping (sigh) the bug

another symptom:

 ./uc machine rm machine-9jfd                                                                               ~miekg/uncloud/cmd/uncloud main
^[[?2026;2$y^[[?2027;2$yNo service containers found on machine 'machine-9jfd'.
This will remove the machine from the cluster and reset it to the uninitialised state.
Do you want to continue? [y/N] y
invalid input. please try again
Do you want to continue? [y/N] y

Machine 'machine-9jfd' removed from the cluster.
Machine reset initiated and will complete in the background.

bubbletea is in no rush to fix this... as an intermediate fix (I tried some source level fixes, but these all failed), I'm now using this function:

uc () {
        if [[ -x ./uc ]]
        then
                ./uc "$@" | cat
        fi
}

which may help someone else too.

I have terrible trouble with Ghostty and random characters appearing while SSH'd in. I've done the infocmp -x xterm-ghostty | ssh YOUR-SERVER -- tic -x - incantation and that at least stops ls from being llss
But I noticed that when I did a do-release-upgrade every time I switched away and back to ghostty there were random characters inserted.
Don't know if that helps ease the pain, thought I'd throw it in the mix.
I'm probably going back to Warp or iTerm2...

shell-integration-features = no-cursor,no-title,ssh-env,ssh-terminfo

takes care of that nowadays

Amazing, thanks. I will stay with ghostty a little longer 😁

got a fix fork here https://github.com/miekg/bubbletea
go mod edit -replace charm.land/bubbletea/v2=github.com/miekg/bubbletea/[email protected] to build uncloud should do the right thing

cli: use forked bubbletea with a fix to drain pending TTY input on sh…

@miekg

@miekg I create a bubbletea fork with a modified fix to drain pending TTY input without the 200ms overhead: https://github.com/unlabs-dev/bubbletea/commits/v2.0.8-uncloud.1/
It's used in main already: 247154e

Can you please test this works for you as well as I can't reproduce this issue in my terminals.

Look ok. But at $WORK cluster access is slightly slower which may not trigger it. Note I tested this in my fork and there it did work beautifully, so this should work as well.

I'll close this for now. Feel free to reopen if the issue persists.

this is still present :(

% uc ls                                                                                                                              ~ limos
^[[?2026;2$y^[[?2027;1$yNAME          MODE         REPLICAS   IMAGE                                         ENDPOINTS
caddy         global       1          caddy/caddy:2026-07-25-182223.1ad0fc5.dirty   (custom Caddy config)
caddy-redis   replicated   1          redis:8.4-alpine
code          replicated   1          codeberg.org/forgejo/forgejo:16               http://code.atoom.net → :3000, http://code.i.atoom.net → :3000
docker        replicated   1          docker:dind
runner        replicated   1          data.forgejo.org/forgejo/runner:12
%
/2027;1$y_

Still an issue on the (currently) latest 0.20.0 for me; can reproduce on Macos with Ghostty 1.3.1 or iTerm2 3.6.11.

looks like those charm folks are working on a AI tui instead of maintaining their stuff. As the docker tui stuff is also brittle it might make sense to move everything over to something more stable (don't know what though)

@miekg

@miekg This fixes the issue for me, feel free to test: #435

fix(cli): stop leaking terminal capability replies into the shell (#435)

Saw it! Thanks!

It's looking good, thanks for tackling this.

Navigation Menu

Repository navigation

[BUG] weird encoding issues on (ghostty) terminals #286

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

[BUG] weird encoding issues on (ghostty) terminals #286

Description

Activity

miekg commented on Mar 24, 2026

miekg commented on Mar 24, 2026

miekg commented on Mar 24, 2026

psviderski commented on Mar 25, 2026

miekg commented on Mar 25, 2026

miekg commented on Mar 25, 2026

miekg commented on Apr 16, 2026

miekg commented on Apr 16, 2026

miekg commented on Apr 21, 2026

rscorer commented on Apr 23, 2026

miekg commented on Apr 23, 2026

rscorer commented on Apr 23, 2026

miekg commented on May 17, 2026

psviderski commented on Jun 17, 2026

miekg commented on Jun 17, 2026

psviderski commented on Jun 18, 2026

miekg commented on Jul 27, 2026

tonyo commented on Sep 5, 2026

miekg commented on Sep 7, 2026

tonyo commented on Sep 27, 2026

miekg commented on Sep 28, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions