It's a bit misleading, but a cool feature of awk....

When the pipe symbol is used awk forks a single process and begins writing to it's stdin or reading from it's stdout depending on where the command and pipe are placed. A more common example is to execute a command and read each record generated:

Code:

function run_it( cmd )
{      while( (cmd|getline)> 0 )   # fork cmd and read from its stdout
        {
           # do something with $0
        }
        close( cmd );
}

As you point out, awk tends to have a limited number of file descriptors available, so it's very important to close them when finished especially in the case of the run_it function that might be called lots of times.

When awk creates the child process, the real FD is mapped using the command string, which is why I tend to put the command into a variable as it's easier to pass it to close, especially if someone comes along and modifies the original command without realising that it needs to be exactly the same in the close.

For the code above, the sort command is forked on the first execution of the statement, and after that awk maps the command string to an already open file descriptor and writes the additional records to the already open FD rather than starting another process. Thus, we get the records we want sorted, and the ones we dont (first and last) are just written to standard output.

This is along the same lines as

Code:

   print foo >"/tmp/file";

Each time the statement is executed the variable contents in foo are printed to the tmp file; the open only happens on the first execution of the statement.

@vgersh99 -- a bit more info than I think you needed, but from the perspective of someone wrestling with the concept of piping a command from inside an awk programme I decided to err on too much.