#fmt should allow any width int, float, etc to be used

[brson]

It would be nice if "%d" would work for any signed integer type, etc. Doing this would require giving extensions access to type information.

[nikomatsakis]

It would be nice if this worked. I wonder if one way would be to simply promote the argument to u64 or i64 as appropriate?

#fmt["%u", x] would translate to u64::to_str(x as u64, 10) etc.

[nikomatsakis]

(Such promotion is, after all, how printf works for types smaller than int)

[brson]

Yeah, we could totally do that. I never really considered implementing it like that because it's a silly little price to pay, but it's very easy, and if there's any real performance impact it's probably not important.

[nikomatsakis]

In general it'd be nice to have type information in the extensions.

[brson]

It would make error messages about incorrect types slightly more opaque (they would be about bad casts to i64, etc.) and would allow for some things that are unexpected, #fmt("%b", 1000) - anything that's castable.

[nikomatsakis]

The point about error msgs is well-taken. As for "%b" (that's boolean, right?), I don't think you need to use as except for the integral types. Still, it'd probably be better to have the actual type info at hand in terms of generating the optimal user experience. But then it'd take 2 seconds to have it working the other way ;)

[brson]

You're right that %b (bool) wouldn't be affected. Basically, I think doing the cast is ok, but if we really wanted to limit the casts to just the obvious stuff, we could add an iface to do the conversion:

mod std::extfmt {

    iface fmt_int<T> {
        fn to_fmt_int() -> i64;
    }

}

Then #fmt could expand every int argument to { import std::extfmt::*; arg.to_fmt_int() }

[nikomatsakis]

I considered suggesting an iface, but I rejected it because it wouldn't be in scope. I didn't think of using a block to limit the scope like that, clever.

[nikomatsakis]

Thinking about this now, I think that if we were to use an iface, we would probably want to have the core imports bring the impls into scope for the built-in types. This would allow people to override them on a file-by-file basis, as well as allowing user-defined types to work with %d, %s, etc. If the aim is not to allow user-defined types and customizable formatting, then I would advocate just casting to u64,i64, etc---it'll compile faster.

[lilyball]

I'd love to see this happen. I'm getting tired of throwing in casts every time I want to print the hex value of a u8. I'd also accept simply adding support for the printf length modifiers, so %hhx would print a u8 as hex.

[ahnurmi]

It seems that currently fmt!("%?", u64::max_value) silently truncates to uint. It would be good to get this fixed, too.