Done!  I added your message to all subcommands, and consolidated some
other checks too.

I really like this change, but I wish it worked for all the cases, not
just the setup subcommand.  That would be kind of annoying to implement
though, so I think this is fine for now.  Maybe I will take a crack at
it!

LGTM!

Stone

I think this is could be a good change, but it won't fix calling message
with a bool.  Currently the definition for message() typechecks the
argument to be a string, so has no need to call coerce_string.

Currently coerce_string is only called during string.format() and
configure_file().  I had the above restriction put in place for these
functions as well, but testing revealed that meson will coerce numbers
and files to strings when substituting values.

That being said, I am happy to merge this, but I'm worried it might have
unintended consequences.

Stone

thank you! applied.

LGTM, thank you!

Stone

I just pushed a fix, which at least works for Chipmunk2D.

Oops!  Very sorry about that.  The reason I changed it back is because
only obj_dependency is handled in the output generator.  Since the
output generator was assuming everything was an obj_dependency, there
were some pretty strange errors.  I suppose the proper fix for this is
to handle obj_external_library in output.c.

By the way, would there be any difference between link_with and
dependency in this case?

Stone

Looks good to me!

I've been thinking for awhile that using uint32_t for everything
(numbers, object ids, string ids, ast nodes) can get confusing.  After
some consideration, I decided on typedefs over other solutions.  To that
end, I've introduced two new typedefs `obj` and `str` and started to
replace all instances of uint32_t.  Until I'm finished, the code might
be even more confusing unfortunately, which is why I thought an email
might be warranted.

Here is some additional information, copied from CONTRIBUTING.md:

> ## Objects, strings, and `uint32_t`
>
> All objects and strings are stored and passed by id, rather than pointer.  Until
> recently, all these ids were `uint32_t`s.  Since that hurts readability, and it