➡️ (and any character with U+FE0F / VS-16) causes TUI misalignment and ghost artifacts

[beadgcf]

Is this a reproducible bug?

• I confirm this is a reproducible bug, not a feature request, idea, question, contribution proposal, or direction check.
• I reproduced this bug on the version and environment reported below using the exact steps provided.

Current behavior

After a pane outputs a character carrying U+FE0F, herdr leaves residual text on screen and the pane border shifts off alignment from that point onward.

Expected behavior

U+FE0F is a zero-width combining mark, so herdr should consume it without advancing the cursor and keep cell accounting in sync — no residual text, no border shift.

Reproduction

Minimal repro from a herdr pane:

printf 'before -> after\n' # plain arrow, no VS-16 — OK
printf 'before ➡ after\n' # U+27A1 alone — OK (text presentation)
printf 'before ➡️ after\n' # U+27A1 + U+FE0F — TUI breaks

Impact

Pane content becomes unreadable whenever an agent's output contains any VS-16 dingbat (common in LLM output — ⚠️, ➡️, etc.).

Environment

• Herdr version: 0.8.0
• Update channel (stable or preview): stable
• Operating system: Windows11 WSL2 Ubuntu 26.04 LTS
• Terminal: Wezterm
• Shell, if relevant: fish
• Relevant config, if any: font Maple Mono NF CN (in wezterm config)

[akbash-bot]

Thanks for the exact repro. One detail will distinguish a WezTerm cell-width mismatch from Herdr's frame serialization. In a plain WezTerm tab outside Herdr, please run this single command and paste its three short output lines:

python3 -c 'exec("import os,re,subprocess,sys,termios,tty\nprint(subprocess.check_output([\"wezterm\",\"--version\"],text=True).strip())\nfd=sys.stdin.fileno(); old=termios.tcgetattr(fd)\ntry:\n tty.setcbreak(fd)\n for name,s in ((\"plain\",\"➡\"),(\"vs16\",\"➡️\")):\n  sys.stdout.write(\"\\r\\x1b[2K\"+s+\"\\x1b[6n\"); sys.stdout.flush()\n  r=b\"\"\n  while not r.endswith(b\"R\"): r+=os.read(fd,1)\n  m=re.search(rb\"\\[(\\d+);(\\d+)R$\",r)\n  sys.stdout.write(\"\\r\\x1b[2K\"+name+\" cursor_column=\"+(m.group(2).decode() if m else repr(r))+\"\\n\"); sys.stdout.flush()\nfinally:\n termios.tcsetattr(fd,termios.TCSADRAIN,old)")'

This only queries the terminal cursor position; it does not collect terminal content or other private data.

[beadgcf]

This command runs in pwsh:

$ wezterm --version
wezterm 20240203-110809-5046fc22

and in wsl bash:

$ python3 -c '
import os,re,subprocess,sys,termios,tty
fd=sys.stdin.fileno(); old=termios.tcgetattr(fd)
try:
 tty.setcbreak(fd)
 for name,s in (("plain","➡"),("vs16","➡️")):
  sys.stdout.write("\r\033[2K"+s+"\033[6n"); sys.stdout.flush()
  r=b""
  while not r.endswith(b"R"): r+=os.read(fd,1)
  m=re.search(rb"\[(\d+);(\d+)R$",r)
  sys.stdout.write("\r\033[2K"+name+" cursor_column="+(m.group(2).decode() if m else repr(r))+"\n"); sys.stdout.flush()
finally:
 termios.tcsetattr(fd,termios.TCSADRAIN,old)
'

plain cursor_column=2
vs16 cursor_column=3

[akbash-bot]

Thanks — those cursor columns confirm WezTerm treats the VS16 grapheme as two cells, matching Herdr’s width model. One final path detail will identify the renderer involved: did you launch the affected TUI as plain herdr, or as herdr --no-session? Please reply with just the exact launch command you used.

[beadgcf]

herdr