An alternative to the pipe (in bash/ksh)

[sophie]

Most people in here will probably know how to pipe output from one command to another:

command 1 | command2

However, what if command2 doesn't allow reading from standard input, and only supports filenames? How can you do this without writing to a file?

It turns out that you can do this in bash and ksh by using the <(command) syntax. For example, the above command can be written:

command2 <(command1)

This will execute command1 in a subshell, and at the same time, call command2 with a file descriptor looking something like /dev/fd/63. When command2 reads from that, it'll get the output of command1.

At first this doesn't seem too useful, but this means that you can do nifty things like this:

diff -u <(sort filea.txt) <(sort fileb.txt)

Which will sort filea.txt and fileb.txt, and then diff the outputs - all without writing a single file.

Note that if the subshells require user input, this isn't going to work, so you can't use this to capture user input and pass it to a script which would otherwise require a filename. However, as long as this isn't the case, everything should work smoothly.

[edit: Oh, and I should mention that, unlike piping, you can execute several commands in the subshell. For example:

rev <(echo wheeness; sleep 2; echo blarg)

This also demonstrates how both the subshell and the main process run simultaneously; the output is "sseneehw", followed by a delay of 2 seconds, followed by "gralb".]

[edit 2: See this comment for an example of how to run multiple commands via piping!]