Leading half-width numerals in Japanese input are dropped from the prompt (e.g., “1人”)

Resolved 💬 8 comments Opened Sep 27, 2025 by genki Closed Oct 22, 2025

💡 Likely answer: A maintainer (tibo-openai, collaborator) responded on this thread — see the highlighted reply below.

What version of Codex is running?

codex-cli 0.42.0

Which model were you using?

gpt-5 high

What platform is your computer?

Darwin 24.6.0 arm64 arm

What steps can reproduce the bug?

When entering Japanese text that starts with a half-width digit, codex-cli drops the leading numeral before sending the prompt. For example, typing ひとり and transformed to 1人 results in 人 being sent.

What is the expected behavior?

The input should be sent unchanged: 1人.

What do you see instead?

The input was changed to: 人

Additional information

_No response_

View original on GitHub ↗

8 Comments

tibo-openai collaborator · 9 months ago

I'm not able to reproduce by either typing or copy pasting the characters above, could you explain a bit more on how to repro this issue?

genki contributor · 9 months ago

@tibo-openai

I have attached the video capturing this issue is really happening.

https://github.com/user-attachments/assets/16b99234-4ef1-4bda-9470-c893a49f45d6

genki contributor · 9 months ago

@tibo-openai

The copy and paste doesn't cause this issue.
Only typing thru IME/Japanese Input transformer causes this issue.
I am using MacOS 15.6.1（24G90）with the default Japanese input method named "日本語入力プログラム".

I have confirmed my PR has solved this issue.

github-actions[bot] contributor · 9 months ago

Potential duplicates detected:

#3260
#2745

Powered by Codex Action

yanoryuichi · 9 months ago

This occurs on both WSL and Windows when using Windows Terminal.
When I enter Japanese characters and delete one character with the Backspace key, two characters disappear from the screen.
It happens not only when the text is entered with the IME, but also when it is pasted from the clipboard. The Japanese characters get garbled or broken.

joshka-oai contributor · 9 months ago

Repro steps

install Japanese - Romanji language and select this
Type hitori (romanji for ひとり)
from the popup select the item that has the half-width character 1人 (may need to tab a few times to select the right part of the popup dialog and search up/down to find this)
press enter a couple of times to insert the characters:

https://github.com/user-attachments/assets/c14bd2c2-497c-451f-b0d4-e4665e5aa34f

Confirmed that the PR fixes this issue:

https://github.com/user-attachments/assets/4b289d0e-97c1-4e2e-8180-a361243b40a3

genki contributor · 9 months ago

@joshka-oai Thank you for confirmation :)

@tibo-openai Would you review my PR?

joshka-oai contributor · 9 months ago

@genki I'm on it right now. Just making sure that the PR doesn't break other parts of the burst paste functionality as we speak and ensuring that the area is a bit better documented than it was to make it easier to understand how some of this code fits together.