Some non-ASCII characters appear as codes like \206 if a key is held down for a long time

[clemente]

What happened?

Holding down keys that produce non-ASCII symbols longer than 2 UTF-8 bytes, like € → …, at some point shows numeric codes instead of the symbol.

Using ./emacs-helm.sh in a recent helm (3e9e33b, from 7th July 2025) with no other patches:

output.mp4

Could it be because garbage collection triggers in while typing?

Apparently it's related, because if I disable GC by setting a high threshold, then the bug doesn't happen:

;; GC never: (setq gc-cons-threshold 100000000000)
;; GC normal: (setq gc-cons-threshold 800000)

How to reproduce?

1. ./emacs-helm.sh, in a TTY under X
2. M-x, or C-x C-f, or any other helm
3. type a specific non-ASCII letter or symbol like → through a key in your keyboard (I hope you have some symbols like that). It doesn't matter if it's pressed together with Shift or with AltGr. See below about the particular symbol to type
4. type it slowly several times, pausing each time. It works
5. Now hold the key down. After around 1 second, it will switch to outputting garbage like \342\206\206\222\206\222\206\222 etc.
6. And if it doesn't, release the key and try step 5 again

What symbol to use

• Letters like á ñ ß ŀ ª ĵ don't trigger this.
• Symbols like → … ∪ ∴ „ ∃ trigger this (yes, I have those symbols accessible through a keypress. I use setxkbmap and a customized keymap).
• And I tried with 明 (which is already a letter, not punctuation) and it also triggers the bug. I mapped it to a key, with symbols[Group2]= [ U660E ]

So I'm assuming that the problem happens with characters that use more than 2 bytes in UTF-8 (like → … ∪ ∴ „ ∃ 明). But those that are 1 or 2 bytes (á ñ ß ŀ ª ĵ) work.

Minor details

• once the numeric codes start appearing, any non-ASCII key will be shown as a code. E.g. á will show \241, ñ will show \261 etc. It's their Unicode codes in decimal
• if I type ASCII letters when the bug is showing, I go back to normal behaviour, and the next ∃ will appear correctly and I can repeat the steps and reproduce the bug again if I want

Where

• Reproduced in urxvt and in Kitty terminals, under X
• It doesn't happen under X, with latest helm, and emacs-gtk from Devuan
• This doesn't happen in the M-x minibuffer in normal TTY emacs without helm
• using Emacs compiled from source, 9f82300bb049c716d59575eaee29262e38c5e8d2, from 10th July 2025

Helm Version

Master branch

Emacs Version

Emacs-30+

OS

GNU/Linux

Relevant backtrace (if possible)

Minimal configuration

• I agree using a minimal configuration

[thierryvolpiatto]

I tried with helm -nw in kitty by pasting continuoulsy "⟶" in M-x or C-x C-f and couldn't reproduce:

[thierryvolpiatto]

Daniel Clemente ***@***.***> writes:

● clemente created an issue (emacs-helm/helm#2730) What happened? Holding down keys that produce non-ASCII symbols longer than 2 UTF-8 bytes, like € → …, at some point shows numeric codes instead of the symbol. Using ./emacs-helm.sh in a recent helm (3e9e33b, from 7th July 2025) with no other patches: https://github.com/user-attachments/assets/6f8db98d-c169-4afd-802c-62721b7b886e Could it be because garbage collection triggers in while typing? Apparently it's related, because if I disable GC by setting a high threshold, then the bug doesn't happen: ;; GC never: (setq gc-cons-threshold 100000000000) ;; GC normal: (setq gc-cons-threshold 800000) How to reproduce? 1. ./emacs-helm.sh, in a TTY under X 2. M-x, or C-x C-f, or any other helm 3. type a specific non-ASCII letter or symbol like → through a key in your keyboard (I hope you have some symbols like that). It doesn't matter if it's pressed together with Shift or with AltGr. See below about the particular symbol to type 4. type it slowly several times, pausing each time. It works 5. Now hold the key down. After around 1 second, it will switch to outputting garbage like \342\206\206\222\206\222\206\222 etc. 6. And if it doesn't, release the key and try step 5 again

Tried now with a TTY (C-M <f1>), log in, type helm, and from M-x type continuously AltGr-i (→) and same couldn't reproduce, I used default gc value.

What symbol to use • Letters like á ñ ß ŀ ª ĵ don't trigger this. • Symbols like → … ∪ ∴ „ ∃ trigger this (yes, I have those symbols accessible through a keypress. I use setxkbmap and a customized keymap).

Did you try without this configuration?

…

• And I tried with 明 (which is already a letter, not punctuation) and it also triggers the bug. I mapped it to a key, with symbols[Group2]= [ U660E ] So I'm assuming that the problem happens with characters that use more than 2 bytes in UTF-8 (like → … ∪ ∴ „ ∃ 明). But those that are 1 or 2 bytes (á ñ ß ŀ ª ĵ) work. Minor details • once the numeric codes start appearing, any non-ASCII key will be shown as a code. E.g. á will show \241, ñ will show \261 etc. It's their Unicode codes in decimal • if I type ASCII letters when the bug is showing, I go back to normal behaviour, and the next ∃ will appear correctly and I can repeat the steps and reproduce the bug again if I want Where • Reproduced in urxvt and in Kitty terminals, under X • It doesn't happen under X, with latest helm, and emacs-gtk from Devuan • This doesn't happen in the M-x minibuffer in normal TTY emacs without helm • using Emacs compiled from source, 9f82300bb049c716d59575eaee29262e38c5e8d2, from 10th July 2025 Helm Version Master branch Emacs Version Emacs-30+ OS GNU/Linux Relevant backtrace (if possible) Minimal configuration • [*] I agree using a minimal configuration — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.*Message ID: ***@***.***>

-- Thierry

[clemente]

Mmmm… I was testing this bug in the latest master, moments before the patch for #2729 was merged. In that version (commit fe2f82f) the bug happens. But after that fix, this problem doesn't happen. You may try that commit.

So that fix also fixed this! Does it make sense? I thought this problem was GC-related and due to the 3-byte UTF-8 codes.
If it makes sense that bug 2729 also fixed this, then please close it, and double-thanks.

Did you try without this configuration?

I just did. I switched to the default Spanish layout (setxkbmap es), where Altgr+E produces the € symbol, which uses 3 bytes. And in the commit id I mentioned above, the bug happens.