Platform CentOS Linux release 7.6.1810, working in bash. GNU bash, version 4.2.46(2)-release (x86_64-redhat-linux-gnu)
This is an idiom I've seen recommended for parsing text in bash in general and in particular for returning multiple values from a function.
IFS=":" read A B <<< $(echo ONE:TWO)
I'm getting unexpected behaviour when I call a function, yyy in the example here
IFS=":" read Y1 Y2 <<< $(yyy)
where yyy itself also wants to do a similar call.
The effect is that that within yyy() even though I explicitly specify the IFS
IFS=":" read C1 C2 <<< $( echo "A:B" )
The fields are parsed, but both values are assigned to C1, it gets the value "A B". If the function is called in isolation it works as expected.
This is a test case, distilled down from a much larger script. I want to know what is happening with IFS here. In the failure case (the second example below) setting IFS=":" in the caller somehow cause the result fields to be aggregated. The first and third calls to yyy() below work as expected, output shown after the code.
#!/bin/bash
debug() { echo "$1" 1>&2 ; }
yyy() {
debug "in yyy"
# why are the two values assigned to A here if the caller specified IFS?
IFS=":" read A B <<< $(echo ONE:TWO)
debug "A=$A"
debug "B=$B"
echo "$A:$B"
}
# this works as expected
read Y1 Y2 <<< $(yyy)
echo -e "===\n"
# this cause the read in yyy() to aggregate
IFS=":" read Y1 Y2 <<< $(yyy)
echo -e "===\n"
# This is a workaround that enables yyy() to work correctly
# But why do I need to do this?
OUT="$(yyy)"
IFS=":" read Y1 Y2 <<< $(echo $OUT)
This is the output
in yyy
A=ONE B=TWO
===
in yyy
A=ONE TWO B=
===
in yyy
A=ONE B=TWO
Note that in the second case A
gets the value ONE TWO
This seems to be a bug in bash-4.2
as discussed here, IFS incorrectly splitting herestrings in bash 4.2. Should work on the versions above that.
These are the results on the same version as you have - GNU bash, version 4.2.46(2). When I ran the function yyy
in debug mode ( by setting set -x
in prompt ).
++ IFS=:
++ read A B
+++ echo ONE:TWO
++ debug 'A=ONE TWO'
++ echo 'A=ONE TWO'
A=ONE TWO
++ debug B=
++ echo B=
B=
++ echo 'ONE TWO:'
The above is snippet of the output from the debug mode output. As you can see when the echo ONE:TWO
is printed as a result of the command substitution, no word splitting is expected to happen because the line doesn't contain any character of the default IFS
value (space/tab or a newline)
So you would expect reading the the whole string with IFS=:
expected to split the string and put the values in the constituent variables A
and B
, but somehow the :
character is lost and a string ONE TWO
is stored as the first variable value.
Look at the output of the function execution in GNU bash, version 4.4.12(1) which exhibits the right behavior.
++ IFS=:
++ read A B
+++ echo ONE:TWO
++ debug A=ONE
++ echo A=ONE
A=ONE
++ debug B=TWO
++ echo B=TWO
B=TWO
++ echo ONE:TWO
There have been lot of IFS
related bugs up to version 4.4.0 bash/CHANGES. So a personal recommendation is to upgrade your bash
version to a more recent stable one. Also see Trying to split a string into two variables
Similar bug on version 4.4.0(1)-release
You would expect the ONE:TWO
to be unmodified when the $(..)
is expanded because for reasons mentioned earlier. But here too the delimit character is lost and the variable A
is set to ONE TWO
IFS=":" read A B <<< $(echo ONE:TWO)
echo "$A"
ONE TWO
Surprisingly the above code works on 4.2.46(2), which means the 4.4.0(1) broke a functionality which used to work in the earlier releases.