Capture and redirect stdin (user input) for runtime of script

Question

I can redirect the stdout/stderr of everything in a script like this:

#!/bin/bash
exec > stdout_file
echo example

But now I want the stdin of the whole script to be written into a file. I tried it like this:

#!/bin/bash
exec 0>stdin_file
sleep 5

If I enter "foo" while the sleep is running it will not be written to the file but bash will try to run the command "foo" after the script exited. I assume this is because bash connects stdin to the stdin of sleep but sleep does not read from stdin.

I want all user input to be redirected to the file, rather than being sent to the stdin of the commands in the script (represented by 'sleep' here).

Among others, I also tried

cat - >stdin_file &

but this doesn't work since bash disconnected cats stdin when sending it to the background.

Answer 1

I'm not sure what your goal is, but trying to get in between the terminal and an interactive command can be somewhat tricky. A tool that's designed to do this is script(1), from util-linux.

Let's say you have a bridge.sh containing:

#!/bin/sh
read -p "What… is your name? " NAME
read -p "What… is your quest? " QUEST
read -p "What… is your favourite colour? " COLOUR
echo "Go on. Off you go."

You can capture the terminal session with:

script -c ./bridge.sh bridge_script

Since you're only interested in the user's input, you can use:

script -c ./bridge.sh --log-in bridge_input_script

The manual page warns:

script places everything in the log file, including linefeeds and backspaces. This is not what the naive user expects.

And indeed, you can see in the input log that there's a bare CR at the end of each of my inputs. (Which makes sense; I hit ⏎, it logged CR.)

> cat -A bridge_input_script
Script started on 2023-05-24 09:46:19-04:00 [TERM="xterm-256color" TTY="/dev/pts/4" COLUMNS="124" LINES="40"]$
My name is Sir Lancelot of Camelot.^MTo seek the Holy Grail.^MBlue.^M$
Script done on 2023-05-24 09:46:32-04:00 [COMMAND_EXIT_CODE="0"]$

(This is really the best it can do without knowing more about the command under instrumentation.)

This is what I usually do, to make sure I have all the data I might need for future analysis:

script -c ./bridge.sh --log-timing bridge_timing_log --log-io bridge_io_log

And I use scriptreplay(1) play back the session:

scriptreplay --log-io bridge_io_log --log-timing bridge_timing_log -m 0.01

(Since I don't care about the timing much, I pass -m 0.01 to cap the maximum delay at 10ms.)

script has a few more tricks up its sleeve, including scriptlive(1), which actually re-runs your session from the logs. (You can really shoot yourself in the foot with this. RTFM and be careful!)

Answer 2

#! /bin/bash -

tee input.log | {
  # contents of your script here
}

Would save stdin to input.log as well as feeding it to the code inside the braces. That code will see the input coming from a pipe though, which may affect its behaviour if it was expecting it coming from a tty device.

You can save both input and output with:

#! /bin/bash -
set -o pipefail
tee input.log | {
  # contents of your script here
} | tee output.log

(pipefail so as not to miss failures of either tee or the code, though beware it will also affect the exit status of pipelines inside the code).