Here is my code:

function update_profile
    echo "1. Update Name"
    echo "2. Update Age"
    echo "3. Update Gender"
    echo "Enter option: "
    read option

    case $option in
         1) update_name ;;
         2) update_age ;;
         3) update_gender ;;

    function update_name
        echo "Enter new name: "
        read name

Just want to make sure if it's possible to do this way. I do know that I can throw all the codes into the case, but it will be messy, so I was thinking to create a stand alone function, within a function, and to be called when needed to perform its commands.


Yes, it's possible.

It is even possible to nest a function within another function, although this is not very useful.

f1 ()

  f2 () # nested
    echo "Function \"f2\", inside \"f1\"."


f2  #  Gives an error message.
    #  Even a preceding "declare -f f2" wouldn't help.


f1  #  Does nothing, since calling "f1" does not automatically call "f2".
f2  #  Now, it's all right to call "f2",
    #+ since its definition has been made visible by calling "f1".

    # Thanks, S.C.

Source: The Linux Documentation Project

  • thinking of using it so that i would not have so much codes in the case, will try your answer, thanks – Zac Jan 17 '15 at 16:35

You can define a function anywhere the shell is expecting a command, including in a function. Note that the function is defined at the moment the shell executes its definition, not when the shell parses the file. So your code won't work if the user chooses option 1 the first time update_profile is executed, because when update_name is called in the case statement, the definition of the function update_name won't have been executed yet. As soon as the function update_profile has been executed once, the function update_name will also be defined.

You need to move the definition of update_name before the point where it is used.

Defining update_name inside update_profile isn't particularly useful. It does mean that update_name won't be defined until the first time update_profile is executed, but it will remain available afterwards. If you want update_name to be available only inside update_profile, define the function inside it and call unset -f update_name before returning from the function. You won't really gain anything by doing that though, compared to doing the simple thing and defining all functions globally.

  • A more simple means of handling the unset ... stuff is just to do fn() ( : ... ; ) – mikeserv Jan 19 '15 at 4:51
  • unset -f update_name is not only simpler than update_name () ( : ... ; ) or the less weird update_name () { :; } but more importantly it does what it's supposed to do, which is to undefine the function. update_name () (…) defines a function which does nothing, so running update_name will do nothing instead of calling the external command by that name. – Gilles Jan 19 '15 at 14:00
  • 1
    weird is a little silly. Moreover, what I meant was: fn() ( fn2() { : define helper; }; do stuff; helper ) where the function is defined to run a subshell. There's nothing weird about that - a function is a named compound command (with new positionals) and not much else. So if a function is defined to run in a subshell from the start - you don't have to clean up its state when it closes. – mikeserv Jan 19 '15 at 14:49

Yes, and this becomes clearer when you consider what a shell function really is.

For POSIX-compliant shells a function definition is standardized thus:

  • 2.9.5 Function Definition Command

    • A function is a user-defined name that is used as a simple command to call a compound command with new positional parameters. A function is defined with a "function definition command"... as follows:

    fname() compound-command[io-redirect ...]

    • The function is named fname... The implementation shall maintain separate name spaces for functions and variables.

    • The argument compound-command represents a compound command, as described in Compound Commands.

      • When the function is declared, none of the expansions in Word Expansions shall be performed on the text in compound-command or <<&io-redirect&>; all ${expansions} shall be performed as normal each time the function is called. Similarly, the optional <<&io-redirect&> redirections and any variable=assignments within compound-command shall be performed during the execution of the function itself, not the function definition. See Consequences of Shell Errors for the consequences of failures of these operations on interactive and non-interactive shells.

And so, at its heart, a shell function named fname is a literal string composed of at least one compound command that the shell will call up from memory and execute in place of fname when it occurs in input in command position - which means wherever a cmd will do. This definition opens a lot of possibilities for the use of a function in a POSIX shell. Either of the following is acceptable:

fn() {
    command; list;
    fn() { : redefines itself upon first execution; }


fn() {
    helper1() { : defines another function which it can call; }
    helper2() { : and another; }
    helper1 "$@" | helper2 "$@"     #processes args twice over pipe
    command "$@"; list;             #without altering its args

But that is a small example. If you consider the meaning of compound command you might begin to see that the conventional fn() { : cmds; } form is only one way a function can work. Consider some different kinds of compound commands:

 { compound; list; of; commands;} <>i/o <i >o
 (subshelled; compound; list; of; commands) <>i/o <i >o
 if ...; then ...; fi <>i/o <i >o
 case ... in (...) ...;; esac <>i/o <i >o
 for ... [in ... ;] do ...; done <>i/o <i >o
 (while|until) ...; do ....; done <>i/o <i >o

And others besides. Any one of the above should work like...

 fname() compound list

...and from among those any that can be nested when not assigned as a function can still be nested even if defined as a command.

Here's one way I might write your function:

    cat >&3
    read "${2-option}" <&3
    case "${1-$option}" in
    1) update_prof '' name      ;;
    2) update_prof '' age       ;;
    3) update_prof '' gender    ;;
    *) unset option             ;;
} <<-PROMPT 3<>/dev/tty
    1. Update Name
    2. Update Age
    3. Update Gender
    Enter ${2:-option}: $(
        printf '\033%s' \[A @

Some notes about the above:

  1. The read in update is subject to IFS and backslash interpretation. Robustly it could be IFS= read -r "$1" but I'm unsure how you wish those things to be interpreted. Look for other answers on this site for more and better information on that score.
  2. The printf '\033%s... in the here-doc assumes /dev/tty is linked to a VT100 compatible terminal, in which case the escapes used should keep the here-doc's final newline from displaying on-screen. Robustly tput would be used. do man termcap for more information there.
    • Best is the VT100 assumption is correct and you can do without either printf or tput by entering the escape characters literally into the here-document like ^V{esc}[A^V{esc}@ where ^V is a way of representing the CONTROL+V key combination and {esc} is your keyboard's ESC key.

The above function will pull double duty depending on its parameter set - it will execute twice and re-evaluate its prompt only as required - and so it doesn't need a second function - because it can do both as long as it is initially called without parameters in the first place.

So if I run...

update_prof; printf %s\\n "$name"

The ensuing terminal activity looks like:

1. Update Name
2. Update Age
3. Update Gender

Enter option: 1

Enter name: yo mama
yo mama
  • 1
    Note that in the Bourne shell (and all other Bourne-like shells except bash, yash and recent versions of posh), you define a function by sticking fname() in front of any command, not just compound ones. Last time I asked, nobody could tell me why POSIX changed that to compound only. bash initially only supported f() { ...; }. – Stéphane Chazelas Jan 20 '15 at 19:23
  • @StéphaneChazelas - no kidding - never even tried that. Just checked it and it works with dash. posh's parser doesn't handle even any compound command though: posh -c 'fname() if :; then echo not broken; fi; fname' results in a syntax error. Also zsh screws up fname() { : ; } <redirect unless the compound form is a subshell. Can I ask though - isn't this then worth an upvote? Or what improvement is necessary to make it so? You might also have missed upvoting your own contribution as quoted in jherran's answer, I think... – mikeserv Jan 20 '15 at 19:32
  • 1
    Though interesting, that's not really answering the question. – Stéphane Chazelas Jan 20 '15 at 19:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.