Difference between ksh and bash script

问题:

Consider the following script (arithmetic syntax that is used for local_var2 is irrelevant for this case):

#!/bin/ksh

function my_func1
{
   typeset local_var1=one
   typeset local_var2
   (( local_var2 = 1 ))
   echo my_func1: local_var1 = $local_var1, local_var2 = $local_var2
}

my_func2()
{
   typeset local_var1=two
   typeset local_var2
   (( local_var2 = 2 ))
   echo my_func2: local_var1 = $local_var1, local_var2 = $local_var2
}
local_var1=0
local_var2=0
echo before functions: local_var1 = $local_var1, local_var2 = $local_var2

my_func1
echo after my_func1: local_var1 = $local_var1, local_var2 = $local_var2

my_func2
echo after my_func2: local_var1 = $local_var1, local_var2 = $local_var2

When run it produces the following output:

before functions: local_var1 = 0, local_var2 = 0
my_func1: local_var1 = one, local_var2 = 1
after my_func1: local_var1 = 0, local_var2 = 0
my_func2: local_var1 = two, local_var2 = 2
after my_func2: local_var1 = two, local_var2 = 2

(which is not what is expected!)

If I run the same script in bash, the output is:

before functions: local_var1 = 0, local_var2 = 0
my_func1: local_var1 = one, local_var2 = 1
after my_func1: local_var1 = 0, local_var2 = 0
my_func2: local_var1 = two, local_var2 = 2
after my_func2: local_var1 = 0, local_var2 = 0

(which is what is expected!)

回答1:

This is one of the weirdness of ksh93.

typesetting of variables to define their scope as local only works with the function definition style:

function func_name
{
}

Not with the function definition style:

func_name()
{
}

With the func_name() style, everything is global. So the behaviour of ksh is as expected!!!

But bash is clearly saner than ksh in this regard. So it sets the scope of variables in both functions as local when typeset was used.

There's FAQ entry in ksh documentation stating the difference between the function defintions:

Q18. What is the difference between function name and name()?

A18. In ksh88 these were the same.  However, the POSIX standard
    choose foo() for functions and defined System V Release 2
    semantics to them so that there are no local variables
    and so that traps are not scoped.  ksh93 keeps the ksh88
    semantics for functions defined as function name, and
    has changed the name() semantics to match the POSIX
    semantics.  Clearly, function name is more useful.

回答2:

Based on the description in this answer, you are most likely running the AT&T version of ksh, for which the typeset built-in makes variables local only in functions declared with the function keyword.