Cfrgtkky

I know I can read parent's information in a child process by exporting a variable

myVariable=1

bash

echo $myVariable



exit

export myVariable=1

bash

echo $myVariable # 1

How can I access the variables defined in the child process from the parent?

bash

a=2

suspend

echo $a

You can't do this.

When you start a new shell, it's an unrelated process and the parent can't "look inside" it to see what it has in memory (this is a security boundary of sorts). Its variables are not shared and not published where someone else can see them.

It gets a copy of the existing environment variables, but any changes it makes to them are for itself alone, and its children. There's no "backwards" communication until the shell exits, and all there is at that point is a single exit code.

When you're doing this interactively as you've shown, that's the end of the story. Your best bet is to save data into a file, which you can then read from or source back in to the parent. If you really can't use a file, I have an awful way of doing it right at the end, but I recommend rethinking what you're trying to achieve - it'd be nice to be able to make this work normally (and some other shells do have shared variables!), it's just not achievable here.

If you're making a program or shell script that runs non-interactively, some or all of source, eval, and printing out data to standard output may help you, as noted in an existing answer. If you exit the child shell, you can pass a single byte of information back in your exit code with exit 123, which you can read with $? in the parent.

If you really really need to migrate something out, here is a very nasty hack to achieve it interactively with no intermediate files:

$ bash

$ export a=2

$ sleep 60 & echo $!

2198

$ suspend

$ xargs -0 < /proc/2198/environ printf '%sn' | grep '^a='

a=2

$ a=$(xargs -0 < /proc/2198/environ printf '%sn' | grep '^a=' | cut -d= -f2-)

$ echo $a

2

$ [1]+  Done                    sleep 60

There are significant limitations here, particularly if you're not in control of the whole environment - a newline within an environment variable value could trick you into setting a value you didn't want, for example.

This works by exporting a so that it appears in the environment of subprocesses and then launching one (sleep 60 here), in the background (&), with the PID of sleep printed out (echo $!). When you suspend the inner shell, you can read the sleep process's environment variables given by its parent from the outside - those are public in the /proc filesystem - but they're separated by null bytes, so xargs -0 splits them up and printf '%sn' outputs one per line. At that point there's a complete list of variables, and we can grep out the one we want, optionally chopping out just the value after the = with cut and setting it. The sleep will end on its own, which is what we see at the end of the transcript ([1]+ Done ...).

You'd need to do it all again if you wanted to pick up on a change to a in the child - and the child will never see a change that you make to a in the parent: all we've done is make another copy. The reason for the sleep process is that we can only see the inbound environment variables of a process, and not any of the shell's internal variable values even if they're exported.

I don't recommend doing this or relying on it, and I can't see any value in using it, but if you do end up in a situation where it's truly, vitally necessary then it should work.

From this question on Stack Overflow the solution is to use source when executing the child process:

In a.sh do source b.sh instead of ./b.sh

a.sh should look like this :

#!/bin/bash

export A=1

source b.sh

echo parent "$A"