Re: Ctrl+C during ~/.bashrc evaluation leads to unexpected behavior
From:
Martin D Kealey
Subject:
Re: Ctrl+C during ~/.bashrc evaluation leads to unexpected behavior
Date:
Tue, 18 Mar 2025 15:46:31 +1000
It's important that sections of ~/.bashrc be ordered correctly to prevent
such issues. Just because someone tells you to add something to the *end*
of your .bashrc does not mean that's necessarily the best place to put it.
1) some boilerplate, including checking whether the shell is actually
interactive bash, plus:
trap '' SIGINT SIGQUIT # ignore ctrl-C and ctrl-\
2) any essential initialization - including variables like HISTFILE,
HISTFILESIZE, HISTSIZE, TERM, TRUECOLOR, TZ and shopt/set options.
3) optionally (useful if the remainder of the file might be "slow"):
trap 'exit $?' SIGINT SIGQUIT # abort login on ctrl-C and ctrl-\
4) read/include (“.”) files:
- recommended by your OS
- mandated by your system administrator
- for any frameworks you want to use (ble, zcomp
- containing your own customisations
(Grouping like this is useful, but some re-ordering may be required to
accommodate interdependencies.)
5) finish with:
trap - SIGINT SIGQUIT # normal handling for ctrl-C and ctrl-\
-Martin
PS: my own preferred approach is to set
HISTFILE=$HOME/.bash_history.d/$EPOCHSECONDS.$$ so that (a) bashrc can't
erase old history, and (b) history from different sessions isn't
interleaved. To recover old history, I use something like:
while read histfile ; do history -r "$histfile" ; done < <( find
~/.bash_history.d/* -mtime -14 -print )
which relies on the `*` glob to sort the histfiles into time order. (I
leave it to my heirs to revise this before 20 November 2286.)
On Tue, 18 Mar 2025 at 05:26, Ionut Nicula <nicula.iccc@gmail.com> wrote:
> Configuration Information [Automatically generated, do not change]:
> Machine: x86_64
> OS: linux-gnu
> Compiler: gcc
> Compilation CFLAGS: -g -O2 -fstack-protector-strong -Wformat
> -Werror=format-security -Wall
> uname output: Linux zinc 6.1.0-32-amd64 #1 SMP PREEMPT_DYNAMIC Debian
> 6.1.129-1 (2025-03-06) x86_64 GNU/Linux
> Machine Type: x86_64-pc-linux-gnu
>
> Bash Version: 5.2
> Patch Level: 15
> Release Status: release
>
> Description:
> Had this weird issue on my work laptop where my .bash_history file
> would get truncated from time to
> time, and I had no idea what caused it because no background or
> foreground process should modify
> .bash_history in that way.
>
> Turns out that this was caused by pressing Ctrl+C when ~/.bashrc
> was being evaluated.
>
> Normally, on my PC, ~/.bashrc is very lightweight and I have a
> very very small window when I can
> press Ctrl+C and have it aborted. But at work, the init takes
> longer (~1 second) and I sometimes
> press Ctrl+C to abort it, run a command that doesn't depend on the
> .bashrc init, and then close
> the shell. But this effectively sets the history length to 500
> (the system default, instead of
> my actual infinite length in ~/.bashrc). When this happened
> before, I didn't notice it right
> away. I only noticed like a couple of hours after I accidentally
> truncated my bash history file,
> so I was very confused. It happened like 8 times in the last 12
> months and it happened again
> today, so I decided to spend a bit of time on it and find the issue.
>
> Repeat-By:
> 1. Write a ~/.bashrc file that sets the history length to allow
> infinite number of entries.
> 2. Write a global bash config that sets the history length limit to
> 500.
> 3. Add 1000 lines to the bash history file.
> 4. Add a command like 'sleep 999' to your ~/.bashrc before the
> lines that set up the infinite
> history size, in order to simulate a slow initialization.
> 5. Start a new bash instance, press Ctrl+C, and notice how
> .bash_history gets truncated to 500
> lines.
>
> Fix:
> I'm not sure what tweaks are 'allowed' here, but ideally, I would
> like some kind of flag that
> simply closes the bash instance if I ever press Ctrl+C during
> .bashrc evaluation because, as my
> report demonstrates, it can quite easily lead to weird and
> confusing behavior. I'm not sure if
> my suggested 'fix' is allowed (or makes sense) to be the default
> behavior.
>
>