How do I tell expect that I have finished the interactive mode?

Question

I am writing some expect commands in bash.

Script:

#!/bin/bash  
set timeout -1  

expect -c "  

spawn telnet $IP $PORT1  
sleep 1  
send \"\r\"  
send \"\r\"  
expect Prompt1>  
interact timeout 20 {  
sleep 1  
}  

expect {  
Prompt2> {send \"dir\r\" }  
}    

"  

My intentions with the script are, first let it telnet into a machine, when it sees Prompt1, let it give control to me, I will execute a command to load a specific image. Then wait until Prompt2 shows up (which indicates image has been loaded). Then Let it execute the further set of commands.

After running the script, I could get into the interactive mode, load my image. The problem is getting out of interactive mode on the remote machine and giving back control to it.

The Error which I got:

expect: spawn id exp4 not open  
    while executing  
"expect -nobrace Prompt2 {send "dir\r" }"  
    invoked from within  
"expect {    
Prompt2 {send "dir\r" }  
}"  

How can I do this?

Answer 1

Your problem is two-fold...

1. You should interact with an explicit return, and give it some way to know you've released control... in this case, I use three plus signs and hit enter.

2. After you return control, the script will need to get the prompt again, which means the first thing you do after returning control to expect is send another \r. I edited for what I think you're trying to do...

Example follows...

#!/bin/bash  
set timeout -1  

expect -c "  

spawn telnet $IP $PORT1  
sleep 1  
send \"\r\"  
send \"\r\"  
expect Prompt1>  
interact +++ return

send \"\r\"
expect {  
Prompt2> {send \"dir\r\" }  
}
"

Answer 2

return = fail

return didn't work for me because in my case, it was not a shell that can simply prompt me again with the same question. I couldn't figure out how to get it to match on what was printed before I did return.

expect_out (to fix above solution) = fail

The manual says:

Upon matching a pattern (or eof or full_buffer), any matching and previously unmatched output is saved in the variable expect_out(buffer).

But I couldn't get that to work (except where I used it below, combined with -indices which makes it work there, and no idea how to make it work to get previous output fed into a new expect { ... } block.)

expect_user

And the solution here using expect_user didn't work for me either because it had no explanation and wasn't used how I wanted, so didn't know how to apply this limited example in my actual expect file.

my solution

So what I did instead was avoid the interactive mode, and just have a way to provide input, one line at a time. It even works for arrow keys and alt+..., (in dpkg Dialog questions) but not for simply <enter> sometimes (hit alt+y for <Yes> or alt+o for <Ok> for those in dpkg Dialog). (anyone know how to send an enter? not '\n', but the enter key like dpkg Dialog wants?)

The -i $user_spawn_id part means that instead of only looking at your spawned process, it also looks at what the user types. This affects everything after it, so you use expect_after or put it below the rest, not expect_before. -indices makes it possible to read the captured part of the regular expression that matches. expect_out(1,string) is the part I wanted (all except the colon).

expect_after {
    -i $user_spawn_id
    # single line custom input; prefix with : and the rest is sent to the application
    -indices -re ":(.*)" {
        send "$expect_out(1,string)"
    }
}

Using expect_after means it will apply to all following expect blocks until the next expect_after. So you can put that anywhere above your usual expect lines in the file.

and my case/purpose

I wanted to automate do-release-upgrade which does not properly support the usual Debian non-interactive flags (see here)...it just hangs and ignores input instead of proceeding after a question. But the questions are unpredictable... and an aborted upgrade means you could mess up your system, so some fallback to interaction is required.