Bash – Shell dropping every other character of input

bashkeyboardrhelsshtmux

I manage software on a bunch of RHEL 6.10 VMs, typically either as a service account or root using sudo -i. Periodically, my shell starts dropping every other character. For example, if I type

cd /usr/local

what actually shows up is

d/s/oa

It's not just cosmetic; if I press enter there, I get

-bash: d/s/oa: No such file or directory

The only solution I've found is to log out of the sudo session (with ^D^D because the first one is ignored) and log back in. This leads me to believe it's not an issue with my terminal or ssh client, but just in case, here's the full stack:

Human > Keyboard > Windows 10 > MobaXTerm > WSL OpenSuse > tmux > ssh > bash > sudo

Obviously, I could try removing tmux from the stack, or WSL altogether, but it would cramp my workflow and be difficult to troubleshoot, as it's an intermittent issue I only encounter about once a week.

What could be causing this?

Best Answer

I have similar keyboard problems with a diferent (and longer) stack of communication programs, like getting invisible characters or rare codes, but I think they are caused by me inadvertently pressing a key combination that changes the terminal functions.

I really don't have time, as you say, for diving in the term definitions looking for which keys could provoke the change, like if pressing a+b+c during a furious typing session could cause the keyboard to send a XOFF ^S command to the remote terminal that stops sending output or something like that. We need a key capturing program to store the keys pressed and another in the remote for getting the received ones to troubleshooting. Also, as there are several programs involved, it is difficult to blame one or other of them.

In the mean time, sometimes a stty sane or reset command in the terminal solved this for me (because of these cases I think is a oscure key term command unknowingly causing the problem); other times I had to restart the shell as you do or even reconnect some portion of the stack.

Related Solutions

Linux kernel dropping custom keyboard scan codes

The kernel sees the weird scan codes and drops them. I would try to get those scan codes values and then update the hardware database index. So in short the plan is this:

get the codes from dmesg output - dmesg should output something like this when the unknown keycode is pressed:
```
Unknown key pressed (translated set 2, code 0xa0 on isa0060/serio0)
```

a0 being the code value.

create custom keycode mapping file. The examples and help are in the default file
(/usr/lib/udev/hwdb.d/60-keyboard.hwdb for Arch, it may be different in other distributions).
update and trigger the hardware database by running the commands:
```
> udevadm hwdb --update
> udevadm trigger /dev/input/eventXX
```

where eventXX corresponds to your keyboard (you can get it by running evtest). You may also reboot instead of triggering.

Look in Arch wiki and the default keycode mapping file for the more detailed description (or in your distribution documentation if it's not Arch).

This is the reliable and simple method, makes mapping on the kernel level so works whatever display server, DE etc is.

Bash – Max character length for Read command (input)

So with bash there are (at least) two options.

The first is read -n 5. This may sound like it meets your needs. From the man page

-n nchars
       read  returns after reading nchars characters rather than
       waiting for a complete line of input, but honor a  delim-
       iter  if fewer than nchars characters are read before the
       delimiter.

BUT there's a gotcha here. If the user types abcde then the read completes without them needing to press RETURN. This limits the results to 5 characters, but may not be a good user experience. People are used to pressing RETURN.

The second method is just to test the length of the input and complain if it's too long. We use the fact that ${#id} is the length of the string.

This results in a pretty standard loop.

ok=0

while [ $ok = 0 ]
do
  echo "Enter the id: " 
  read id
  if [ ${#id} -gt 5 ]
  then
    echo Too long - 5 characters max
  else
    ok=1
  fi
done

If you want it to be exactly 5 characters then you can change the if test from -gt to -eq.

Best Answer

Related Solutions

Linux kernel dropping custom keyboard scan codes

Bash – Max character length for Read command (input)

Related Question