How can function definition be part of pipe sequence in POSIX shell grammar?

Question

The POSIX shell grammar at http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#tag_18_10_02

says

pipe_sequence    :                             command
                 | pipe_sequence '|' linebreak command
                 ;
command          : simple_command
                 | compound_command
                 | compound_command redirect_list
                 | function_definition

which means, function definition can be a term in a pipe sequence. How is this possible? The function definition cannot have standard input or output and it is not a command that can be executed. Only a function call, which is a simple command, can be executed.

Added after the first comment and answer:

If we split off function_definition from command here, and add it as another alternative wherever else command appears, then yes we are complicating the grammar a little.

But the payoff is much more important: the implementation of such shell, is much easier.

Because if you allow function definitions in a pipe, you have to deal with questions such as what is the scope of the function, and in what environment does it run. I don't believe such questions are in fact answered in the standard at all.

What is worse: a little more complexity in the grammar, or much more work and complexity for the implementer. If the former, then is this not a case of "tail wagging the dog"?

Answer 1

It's actually easier for the implementer to not have to worry about this. When doing a pipe each component is run in its own subshell (except maybe the first in bash, or the last in ksh88/ksh93 if the command is a native one). Thus the function definition in the middle of a pipeline would be defined for the shell instance for that component of the pipe, but not visible outside... and this is all automatic based on the semantics of pipelines.

If you wanted to prevent function definitions (or alias definitions, or silly commands such as cd...) inside a pipeline then you've complicated the implementation :-)

Answer 2

Why wouldn’t it be possible? Is it pointless? Definitely. But it works:

$ function asdf { echo "bla"; } | hexdump -C; echo EOF
EOF

Similarly:

$ function asdf { echo "bla"; } | asdf | hexdump -C; echo EOF
-bash: asdf: command not found
EOF

Defining a function is a “command” like any other. It doesn’t have any output and doesn’t take any input, though. You could even do a variable assignment. Pointless again, of course, but not an error.

The “why” is probably: KISS. You wouldn’t want to pollute your grammar with needless complexity.

Update: Upon further examination I found out that Bash doesn't even bother running the pipe commands after a function definition.