Frequently Asked Questions

Unicode glyphs render as underscores in my tmux!

This is likely an issue with LANG and locale. tmux will substitute unicode glyphs with underscores if it believes that your environment doesn't support UTF-8.

If you're running on macOS, upgrade to 20200620-160318-e00b076c or newer and WezTerm will automatically set LANG appropriately.

Note that if you change your environment you will likely need to kill and restart your tmux server before it will take effect.

You probably should also review this relevant section from the TMUX FAQ, and read on for more information about LANG and locale below.

Some glyphs look messed up, why is that?

There's a surprisingly amount of work that goes into rendering text, and if you're connected to a remote host, it may span both systems. Read on for some gory details!

LANG and locale

Terminals operate on byte streams and don't necessarily know anything about the encoding of the text that you're sending through. The unix model for this is that the end user (that's you!) will instruct the applications that you're running to use a particular locale to interpret the byte stream.

It is common for these environment variables to not be set, or to be set to invalid values by default!

If you're running on macOS, upgrade to 20200620-160318-e00b076c or newer and WezTerm will automatically set LANG appropriately.

You need to select a unicode locale for best results; for example:

export LANG=en_US.UTF-8
# You don't strictly need this collation, but most technical people
# probably want C collation for sane results

If you have other LC_XXX values in your environment, either remove them from your environment (if applicable) or adjust them to use a UTF-8 locale.

You can run locale -a to list the available locales on your system.

You need to make sure that this setting applies both locally and on systems that you log in to via ssh or the mux connection protocol.

If you're seeing multiple garbage characters in your terminal in place of what should be a single glyph then you most likely have a problem with your locale environment variables.

Pasting or entering unicode in zsh looks broken

By default, zsh's line editor doesn't support combining character sequences. Make sure that you have LANG and local configured correctly as shown above, and then tell zsh to enable combining characters:


You'll want to put that into your zshrc so that it is always enabled.

See this stackexchange question for more information.

Fonts and fallback

If you have configured the use of a font that contains only latin characters and then try to display a glyph that isn't present in that font (perhaps an emoji, or perhaps some kanji) then wezterm will try to locate a fallback font that does contain that glyph.

Wezterm uses freetype and harfbuzz to perform font shaping and rendering in a cross platform way, and as a consequence, doesn't have access to the system font fallback selection. Instead it has a short list of fallback fonts that are likely to be present on the system and tries to use those.

If you're seeing the unicode replacement character, a question mark or in the worst cases spaces where a glyph should be, then you have an issue with font fallback.

You can resolve this by explicitly adding fallback font(s) the have the glyphs that you need in your .wezterm.lua:

local wezterm = require 'wezterm';

return {
  font = wezterm.font_with_fallback({
    "My Preferred Font",
    -- This font has a broader selection of Chinese glyphs than my preferred font

See also Troubleshooting Fonts.

Some (but not all) Emoji don't render properly

To some extent this issue can manifest in a similar way to the LANG and locale issue. There are different versions of the Emoji specifications and the level of support in different applications can vary. Emoji can be comprised from a sequence of codepoints and some combine in interesting ways such as a foot and a skin tone. Applications that don't support this correctly may end up emitting incorrect output. For example, pasting some emoji into the zsh REPL confuses its input parser and results in broken emoji output. However, if you were to emit that same emoji from a script, wezterm would render it correctly.

If you're seeing this sort of issue, then you may be able to upgrade the affected application on that system to see if a newer version resolves that issue.

Multiple characters being rendered/combined as one character?

wezterm supports advanced font shaping, which, amongst other things, allows for multiple characters/glyphs to be combined into one ligature. You may be experiencing this if, e.g., != becomes rendered as in wezterm.

If you are seeing this kind of "font combining" and wish to disable it, then this is documented in advanced font shaping options page.

How to troubleshoot keys that don't work or produce weird characters!?

There are a number of layers in input processing that can influence this.

The first thing to note is that wezterm will always and only output UTF-8 encoded text. Your LANG and locale related environment must be set to reflect this; there is more information on that above.

If the key in question is produced in combination with Alt/Option then this section of the docs describes how wezterm processes Alt/Option, as well as options that influence that behavior.

The next thing to verify is what byte sequences are being produced when you press keys. I generally suggest running xxd, pressing the relevant key, then enter, then CTRL-D. This should show a hex dump of the the byte sequence. This step helps to isolate the input from input processing layers in other applications.

Interactive Unix programs generally depend upon the TERM environment variable being set appropriately. wezterm sets this to xterm-256color by default, because wezterm aims to be compatible with with the settings defined by that terminfo entry. Setting TERM to something else can change the byte sequences that interactive applications expect to see for some keys, effectively disabling those keys.

On top of this, a number of programs use libraries such as GNU readline to perform input processing. That means that settings in your ~/.inputrc may changing the behavior of bash. Verify any settings in there that might influence how input is resolved and see the question below about convert-meta!

If you are using tmux be aware that it introduces its own set of input/output processing layers that are also sensitive to LANG, TERM and locale and how they are set in the environment of the tmux server when it was spawned, the tmux client and inside the processes spawned by tmux. It is generally best to troubleshoot input/output weirdness independent of tmux first to minimize the number of variables!

If after experimenting with your environment and related settings you believe that wezterm isn't sending the correct input then please open an issue and include the xxd hexdump, and output from env and any other pertinent information about what you're trying and why it doesn't match your expectations.

I have set convert-meta on in my ~/.inputrc and latin characters are broken!?

That setting causes Readline to re-encode latin-1 and other characters as a different sequence (eg: £ will have the high bit stripped and turn it into #).

You should consider disabling that setting when working with a UTF-8 environment.

How do I enable undercurl (curly underlines)?

Starting in version 20210314-114017-04b7cedd, WezTerm has support for colored and curly underlines.

The relevant escape sequences are:

 CSI 24 m   -> No underline
 CSI 4 m    -> Single underline
 CSI 4:0 m  -> No underline
 CSI 4:1 m  -> Single underline
 CSI 4:2 m  -> Double underline
 CSI 4:3 m  -> Curly underline
 CSI 4:4 m  -> Dotted underline
 CSI 4:5 m  -> Dashed underline

 CSI 58:2::R:G:B m   -> set underline color to specified true color RGB
 CSI 58:5:I m        -> set underline color to palette index I (0-255)
 CSI 59              -> restore underline color to default

You can try these out in your shell; this example will print the various underline styles with a red underline:

$ printf "\x1b[58:2::255:0:0m\x1b[4:1msingle\x1b[4:2mdouble\x1b[4:3mcurly\x1b[4:4mdotted\x1b[4:5mdashed\x1b[0m\n"

To use this in vim, add something like the following to your .vimrc:

let &t_Cs = "\e[4:3m"
let &t_Ce = "\e[4:0m"
hi SpellBad   guisp=red gui=undercurl guifg=NONE guibg=NONE \
     ctermfg=NONE ctermbg=NONE term=underline cterm=undercurl ctermul=red
hi SpellCap   guisp=yellow gui=undercurl guifg=NONE guibg=NONE \
     ctermfg=NONE ctermbg=NONE term=underline cterm=undercurl ctermul=yellow

If you are a neovim user then you will need to install a terminfo file that tells neovim about this support.

You may wish to try these steps to install a copy of a wezterm terminfo file; this will compile a copy of the terminfo and install it into your ~/.terminfo directory:

tempfile=$(mktemp) \
  && curl -o $tempfile \
  && tic -x -o ~/.terminfo $tempfile \
  && rm $tempfile

With that in place, you can then start neovim like this, and it should enable undercurl:

env TERM=wezterm nvim

Note: on Windows, the ConPTY layer strips out the curly underline escape sequences. If you're missing this feature in your WSL instance, you will need to use either wezterm ssh or multiplexing to bypass ConPTY.

I use Powershell for my shell, and I have problems with cursor keys in other apps

Powershell has an open issue where it enables the DECCKM mode of the terminal and does not restore it prior to launching external commands.

The consequence of enabling DECCKM is that cursor keys switch from being reported as eg: ESC [ A (for UpArrow) to ESC O A.

Some applications don't know how to deal with this and as a consequence, won't see the cursor keys.

This is not an issue in WezTerm; the same issue manifests in any terminal emulator that runs powershell.