Krdytkyu

For a short period of time, I want to save the command history and output of every command I run in any terminal I open. It should all go in the same file.

So basically I want to run a command (or edit bashrc) such that I can then;

• Open terminal #1 and run echo hello
  


• Open terminal #2 and type ls
  


• Close terminal #1 and open terminal #3 and type whoami
  


• Close all terminals. Open terminal #4, and type a command to see all the commands run in order above (echo hello, ls, whoami), plus their output, in the order I ran the commands.

I could in theory open each terminal and type

screen -f output.txt

and then

exit

before I closed each terminal, but I don't want to have to remember to type that every time. I just want it done automatically until I manually stop it (by either running a command, or updating bashrc).

Please read the question carefully before declaring it a duplicate. I have searched existing questions and did not find something equivalent (though its possible I missed it).

Potential solution

After some research and experimenting, here's what I came up with:

logstuff(){



    while true; do



        case $1 in

            "on" ) exec > >( ( printf ">>>>> TIME:$(date) SHELLPID:$$;n"; tee /dev/tty ; printf ">>>>>n" ) >> logfile.txt) 2>&1 ;

                   break;;

            "off") exec > /dev/tty 2>&1 ;

                   break;;

                *) echo "Please type 'on' or 'off';;

        esac

    done

}

This bash function should be placed in your ~/.bashrc and is available for use when opening new terminal or after issuing source ~/.bashrc. Logging has to be turned on manually via on and off arguments.

DEMO

Here's how it works in practice:

Do stuff in shell 1:

<shell 1>$ logstuff on

<shell 1>$ stat /etc/passwd

  File: /etc/passwd

  Size: 2208            Blocks: 8          IO Block: 4096   regular file

Device: 802h/2050d      Inode: 156236      Links: 1

Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)

Access: 2018-12-01 20:24:02.620000000 +0000

Modify: 2018-10-31 01:33:42.701000999 +0000

Change: 2018-10-31 01:33:42.704998999 +0000

 Birth: -

<shell 1>$ logstuff off

<shell 1>$ 

Do stuff in shell 2:

< shell 2 >$ logstuff on

< shell 2 >$ echo "Hello World !"

Hello World !

< shell 2 >$ logstuff off

< shell 2 >$ 

Now review the logfile.txt:

<shell 1>$ cat logfile.txt 

>>>>> TIME:Sat Dec  1 21:43:00 UTC 2018 SHELLPID:2225;

<shell 1>$ stat /etc/passwd

  File: /etc/passwd

  Size: 2208            Blocks: 8          IO Block: 4096   regular file

Device: 802h/2050d      Inode: 156236      Links: 1

Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)

Access: 2018-12-01 20:24:02.620000000 +0000

Modify: 2018-10-31 01:33:42.701000999 +0000

Change: 2018-10-31 01:33:42.704998999 +0000

 Birth: -

<shell 1>$ >>>>> TIME:Sat Dec  1 21:43:11 UTC 2018 SHELLPID:2359;

< shell 2 >$ echo "Hello World !"

Hello World !

< shell 2 >$ logstuff off

>>>>>

logstuff off

>>>>>

<shell 1>$ 

Issues

• If logstuff on is issued in both terminals first, there's a chance of outputs being mangled together. The way it works is that you have to issue logsutff on in shell 1, then issue commands there, then issue logstuff on in shell 2.


• This uses process substitution >( ), tee and a subshell. Not the most elegant nor efficient due to bunch of forking and extra pipeline.


• 
  logfile.txt is stored in current working directory. This should be changed to ~/logfile.txt or however the user sees fit.

Practical considerations

What the question itself asks is somewhat impractical: storing output from multiple shells into one single file means you're storing output of commands from two or more completely unrelated session, which may have different environment variables, different working directories, or working on different filesystems; this means there's whole lot of context lacking if you're intending to use such log text for debugging purposes or solving a problem.

Far better approach would be to have script -f write to log files in one specific directory, potentially with filenames of such logs timestamped or appended the shell PID. Another solution - instead of having 3 different terminals, just use one - screen or my personal favorite byobu-screen. You can attach/detach to a single virtual tty session in screen and it is often used to keep processes running on remote servers where you have to log out but still need a shell session with output and tracebacks running. This can be combined with script as well.