Support shell()

[gyreas]

This PR add support for the shell(), which expands to the output of running its given command.

Example:

set shell := ["fish", "-c"]
v := shell("
set p $argv
for i in $p
	echo "$i"
end
",  "Nightmares", "on", "Elms", "Street")

_:
  @echo "{{ v }}"

output:

Nightmares
on
Elms
Street

[casey]

This seems really nice! What do you think about getting rid of the name argument for now? Since a lot of uses of this function won't use it, it seems unfortunate to force users to pass an additional dummy argument if they want to pass arguments. I'd like to enable it in the future with a dedicated named argument syntax, i.e., shell(CMD, name=NAME, ARGS), which is generally useful and can be done in a future PR.

[casey]

Also, this should be documented in the readme.

[gyreas]

Alright. I clean stuff up, I think I have a solution to remove that inconvenience. Thanks

[casey]

I think this should be refactored to use the same code as Evaluator::run_backtick. run_backtick does a few additional things, like setting the working directory, exporting environment variables, and the destination of stderr.

To make this easier, I just landed #2048, which gets rid of FunctionContext and instead just passes the Evaluator, since all fields on FunctionContext are just taken from Evaluator anyways.

The signature of run_backtick can be changed to something like:

fn run_command(&self, command: &str, args: &[String], token: Option<&Token<'src>>) -> RunResult<'src, String> {

Which produces an Error::Backtick when token is present, and an Error::FunctionCall when it isn't.

[casey]

Nice, LGTM!

[gyreas]

Thank you!

[casey]

Thank you! shell() is a great addition!