interface around {fmt} for cpp11::stop() and cpp11::warning()` message formatting

[sbearrows]

@jimhester Let me know what you think! Tests are pretty simple but we can add more if issues arise.

Fixes #169

[jimhester]

So I was thinking about this some this morning, and I think we will either have to use new function names (something like stop_fmt() for the fmt powered functions, or put them behind a #ifdef CPP11_USE_FMT flag that you have to opt into to use them.

This is because existing packages that are using cpp11::stop() are going to be using the sprintf style %s etc. message strings, and these won't work with fmt.

Another argument in favor of the #ifdef approach is that compiling the fmt code will incur some overhead, which some people may want to avoid.

Which way do you think would be best?

[sbearrows]

I'd be in favor of the later! I think decreasing overhead is probably important to a lot of people using cpp11. What about you?

[jimhester]

yeah that is the way I was leaning, though I worry if it is only opt-in then people may not realize it is an option / never use it.

[sbearrows]

We could try to make a ruckus about? Adding examples in documentation?

Also, could we make fmt for cpp11::stop() and cpp11::warning() opt-in but make fmt for the new cpp11::messages() opt-out since it isn't present in previous versions? This might give people some breadcrumbs for the others? Not sure if that would be too confusing though.

[bkietz]

Also, it's unfortunate that we're losing the %x format specifier for SEXPs. This could be recoverable by introducing a helper function:

// pass through by default
template <typename T>
const T& as_formattable(const T& value) { return value; }

// specialization for SEXP: 
std::string as_formattable(SEXP sexp) { return gettextf("%x", sexp); }

Used like so:

Suggested change

	std::string msg = fmt::format(fmt_arg, args...);
	std::string msg = fmt::format(fmt_arg, as_formattable(args)...);

[jimhester]

I think you would use cpp11::stop("{0:x}", my_sexp), or ({0:#x} if you want the leading 0x) to print the hexadecimal address for a sexp. (https://fmt.dev/latest/syntax.html#format-examples)

[bkietz]

Oh, I misunderstood the purpose of %x. I thought it gave a string representation of the pointed value rather than a representation of the pointer itself. Let me rephrase: would it be desirable to intercept SEXP pointers and replace them with capture.output(print(value)) or so? It'd probably be better to include that as a sexp::to_string() so that we can write

cpp11::stop("Invalid: {}", cpp11::sexp(value).to_string());

[jimhester]

Ah, maybe, though it is not always clear what output would be preferred for a given SEXP, possibly something like that returned from R_inspect() might be the most useful for any SEXP type, though the default arguments can generate a lot of data for deeply nested objects.

> .Internal(inspect(1:10))
@7ff6163c1388 13 INTSXP g0c0 [REF(65535)]  1 : 10 (compact)

[jimhester]

LGTM, I think this can be merged once the tests are green.