howto get exit code of script running within expect

Question

for some time I am struggling to get the exit code of a script which I am running from within expect. It is a BASH script and the questionable part looks like this:

  expect -c "
    log_file $LOG
    spawn su - $ora_user
    expect ""
    send \"source $oraenv_binary\r\"
    expect \"ORACLE_SID = \[$ora_user\] \?\"
    send \"$SID\r\"
    expect \"The Oracle base has been set to /oracle/$SID\"
    send \"$execPATHroot/$subscript $args_subscript\r\"
    expect ""
    send \"echo \$?\r\"
    expect -re \"(\\d+)\" {
    set result \$expect_out(1,string)
    }
    send_user \"subscript exit code: \$result\"
    log_file
    send \"exit\r\"
    expect ""
    exit [lindex \$result 3]"
    sub_rc=$?

Needed to say that this is one of many tries to get the code, however, unsuccessfully. I guess that my problem lies in incorrectly escaped characters or wrong use of brackets.....

When debugging, I am getting the following:

[336] oraenv_binary=/usr/local/bin/oraenv
[338] expect -c '
    log_file /var/opt/osit/oracle/log/ora_sbp_patching_root.bash.log
    spawn su - oracle
    expect
    send "source /usr/local/bin/oraenv\r"
    expect "ORACLE_SID = \[oracle\] \?"
    send "H95\r"
    expect "The Oracle base has been set to /oracle/H95"
    send "/opt/osit/oracle/bin/ora_sbp_patching_orausr.bash -s H95 -a CHECK -p /imports/e2r2s48ifde0002/CDSAP/DB/oracle/ORA19/SBP/SBP_1915_220419_202205 -h /imports/e2r2s48ifde0002/CDSAP/DB/oracle/ORA19/SBP/SBP_1915_220419_202205/README19P_2205-70004508.HTM -u oracle\r"
    expect
    send "echo $?\r"
    expect -re "(\d+)" {
    set result $expect_out(1,string)
    }
    send_user "subscript exit code: $result"
    log_file
    send "exit\r"
    expect
    exit [lindex $result 3]'

.....subscript runs here OK with exit code 0 in this case

-sh-4.2$ subscript exit code: decho $?
0
-sh-4.2$ exit
logout
expected integer but got ""
    while executing
"exit [lindex $result 3]"
[357] sub_rc=0

It seems to me that the regex part "(\d+)" is not OK, but perhaps, it is completely a mess... :-) Please help.

I have read and tried these recommendations: Is there a way to expect output without removing it from the buffer in Tcl expect?

https://wiki.tcl-lang.org/page/How+Expect+can+capture+the+exit+code+from+a+remote+command

https://www.unix.com/shell-programming-and-scripting/144812-expect-script-obtain-exit-code-remote-command.html

Answer 1

You have this

    expect -re "(\d+)" {
    set result $expect_out(1,string)
    }
    send_user "subscript exit code: $result"

and we can see the output is

-sh-4.2$ subscript exit code: decho $?
0

Because the regular expression "(\d+)" is in double quotes , backslash substitutions will occur, and the pattern becomes (d+) which may not match (do you get a 10 second delay at that point?) -- I suspect this is why $result is empty.

Backslashes are prevalent in regular expressions. Using braces to quote them is the way to go:

expect -re {\d+} {set result $expect_out(0,string)}

---

Running your expect code with expect -d (or set exp_internal 1 in the code) emits very verbose expect debug output that is extremely useful to see how your patterns are matching (or not).

---

Using quoted shell heredocs is (IMO) preferable to using quoted strings to encapsulate code.

Consider

expect -c "
send_user \"my home is \$env(HOME)\\n\"
"

versus

expect << 'END_CODE'
    send_user "my home is $env(HOME)\n"
END_CODE

With this technique, you pass shell variable to expect through the environment:

export ora_user=oracle

expect << 'END_EXPECT'
    #...
    spawn su - $env(ora_user)
END_EXPECT

Answer 2

thanks a lot for your answer Glenn, interesting points mentioned. Regarding the braces versus double quotes - I have changed it like that, but no effect. I have those double quotes escaped by backslash within my code - I think the effect is the same, however, definitely it looks nicer to me and evidently it is safer. I have played with the debug mode of expect - thanks for that, I can see much more info. I have noticed that expect holds much more stuff than I "expected":-)

==> this is just a snippet:

.\r\n-sh-4.2$ " (spawn_id exp7) match regular expression "\d+"? (No Gate, RE only) gate=yes re=yes
expect: set expect_out(0,string) "4"
expect: set expect_out(spawn_id) "exp7"
expect: set expect_out(buffer) "\r\n-sh-4"
can't read "expect_out(1,string)": no such element in array
    while executing
"set result $expect_out(1,string)"
    invoked from within
"expect -re {\d+} {set result $expect_out(1,string)}"

As you can see, when I am sending the subscript to be executed I am expecting "" ie nothing, just the new prompt line. However, at that point expect is full of stuff, not at all blank - I think, I need to define the prompt exactly:

-sh-4.2$

and then I need to expect it, together with echoed exit code $? and somehow separate exit code integer to get what I want...... I will keep trying.

Answer 3

The final code which works is as follows (much of the credit goes to Glenn for his numerous advices):

expect -c "
log_file $LOG
spawn su - $ora_user
expect -re {\$ $}
send \"PS1='>'\r\"
expect -re {>$}
send \"source $oraenv_binary\r\"
expect {ORACLE_SID = \[$ora_user\] ? }
send \"$SID\r\"
expect \"The Oracle base has been set to /oracle/$SID\"
send \"$execPATHroot/$subscript $args_subscript\r\"
expect { (subscript) ; exp_continue }
expect -re {>$}
send \"echo \$?\r\"
expect -re {(\d+)\r\n>$} {set result \$expect_out(1,string)}
send_user \"subscript exit code:\$result\"
log_file
send \"\rexit\r\"
expect \"logout\"
exit [lindex \$result 0]"
sub_rc=$?
echo sub_rc:$sub_rc

The first thing after spawn su - $ora_user is to set the prompt by send \"PS1='>'\r\" in order to make new lines with prompt less intrusive to expect.

Then after send \"$execPATHroot/$subscript $args_subscript\r\" I have used the fact, that I have written the subscript to have every line of output populated by (subscript) keyword. So while the subscript produces the output, the expect keeps going by exp_continue .

When the $subscript ends, the prompt > appears into which the expect sends echo $? to get exit code of the $subscript . This appears on the screen as:

>echo $?
0
>

so the code should expect the integer, return and the new line with prompt - ie {(\d+)\r\n>$} . At that time the expect matches the output and expect_out(1,string) is correctly populated:

send: sending "echo $?\r" to { exp7 }
Gate keeper glob pattern for '(\d+)\r\n>$' is '*
>'. Activating booster.
expect: does "" (spawn_id exp7) match regular expression "(\d+)\r\n>$"? Gate "*\r\n>"? gate=no
echo $?
0
>
expect: does "echo $?\r\n0\r\n>" (spawn_id exp7) match regular expression "(\d+)\r\n>$"? Gate "*\r\n>"? gate=yes re=yes
expect: set expect_out(0,string) "0\r\n>"
expect: set expect_out(1,string) "0"
expect: set expect_out(spawn_id) "exp7"
expect: set expect_out(buffer) "echo $?\r\n0\r\n>"

Aother thing to mention is \r within send \"\rexit\r\" as the preceding send_user \"subscript exit code:\$result\" does not result in new line and prompt.

The last change to the code in question is:

exit [lindex \$result 0]"

I have changed the index to 0 as variable result has just one item and index 0 stands for 1st item in the list.