Don't throw away unused arguments of format_args

[m-ou-se]

Fixes #115423

format_args!("{}", 12345u8) was accidentally accepted, even though that literal is out of range. The literal didn't end up in the expansion of format_args!(), because it got simplified to format_args!("12345").

This change is simply to not throw away those 'inlined' arguments. This actually simplifies the code in several ways. :)

[rustbot]

r? @WaffleLapkin

(rustbot has picked a reviewer for you, use r? to override)

[m-ou-se]

I just realized that this has a subtle bug, so don't merge this yet. :)

[WaffleLapkin]

Have you seen #116633? Is this a replacement of that PR?

[m-ou-se]

Ah, I didn't see that PR. Oops.

That PR fixes the issue too. (It seems to be stuck though?)

There's a subtle difference, and that is that with #[allow(overflowing_literals)], print!("{}", 1234u8) with this PR will still result in 1234, while that PR will result in 210, which would be more correct.

I really don't understand why overflowing literals is a lint rather than a hard error though.

[WaffleLapkin]

Yeah, given the T-lang approval I really wish the result is 210. What do you think should be the process here? You change this PR? We ping the author of the other PR? Something else?

[WaffleLapkin]

@rustbot author

[bors]

☔ The latest upstream changes (presumably #119324) made this pull request unmergeable. Please resolve the merge conflicts.

[m-ou-se]

Closing in favor of #116633.