Process substitution

<( <LIST> )

>( <LIST> )

This only works on systems that support named pipes (FIFO - a special file) or the /dev/fd/NNNNN-syntax to access open files.

Process substitution is performed simultaneously with parameter expansion, command substitution and arithmetic expansion.

The command list <LIST> is executed and its

• standard output filedescriptor in the <( ... ) form or
• standard input filedescriptor in the >( ... ) form

is connected to a FIFO or a file in /dev/fd/. The filename (where the filedescriptor is connected) is then used as a substitution for the <(...)-construct.

That, for example, allows to give data to a command that can't be reached by pipelining (that doesn't expect its data from stdin but from a file).

This code is useless, but it demonstrates how it works:

$ echo <(ls)
/dev/fd/63

The output of the ls-program can be accessed by reading the file /dev/fd/63.

This is very useful in redirection statements. A wrong piece of code to count all files in /etc is:

counter=0

find /etc | while read; do
  ((counter++))
done

echo "$counter files"

Why this is wrong? Due to the pipe, the while read; do ... done part is executed in a subshell. That also means counter is incremented in that subshell. When the pipeline finishes, the subshell is terminated, and the counter you access is the counter from the main shell - the one at "0"!

Process substitution helps us here to avoid the pipe (which is the reason for the subshell):

counter=0

while read; do
  ((counter++))
done < <(find /etc)

echo "$counter files"

This is the normal input file redirection < FILE, just that FILE is dynamically generated by process substitution:

< <(COMMAND)

• Internal: Introduction to expansion and substitution
• Internal: Bash in the process tree (subshells)
• Internal: Redirection