Add diagnostics for E0172, fix inline message for E0139

[AlisdairO]

As title :-)

Part of #24407.

[Manishearth]

So sorry, I already did this in Manishearth@850e9be .

I'll update the spreadsheet to give myself a global lock over librustc

[Manishearth]

If you think that one can be improved or this one is better, let me know and we can merge them :)

[AlisdairO]

Ah that's a shame :(. Oh well - is E0172 still okay? I could pull that out into a separate PR.

On 139, I think yours is a better explanation of more common scenarios to hit this problem. That said, I think the short error message may be a little off? It talks about type parameters in its interior, but for the following example, I wouldn't really call the parameter 'interior':

fn foo<T: std::fmt::Display()>(bar: T) {
    unsafe {
        let baz: std::raw::TraitObject = std::mem::transmute(bar);
    }
}

[Manishearth]

Sure, in fact just amending the commit in this PR would work too.

Hmm. "Interior" might be alluding to situations where there are no explicit type params involved.

E.g.

fn foo<T>(x: *const T) -> Box<T>{
type ConcreteRaw = *const T;
type Concrete = Box<T>;
transmute::<ConcreteRaw, Concrete>(x);
}

But this isn't an important distinction, really. "contaisn unsubstituted type parameters" or some such might be better. (Feel free to make the change to the inline error in this same PR)

[AlisdairO]

OK, I think that should do it!

[Manishearth]

nit: backticks around Display

[Manishearth]

LGTM modulo nit

[AlisdairO]

Added the nit-fix, and squashed.

[Manishearth]

@bors r+ rollup

[bors]

📌 Commit 686d326 has been approved by Manishearth

[GuillaumeGomez]

@AlisdairO: If you wanna see what error code remains, take a look at the spreadsheet available here : #24407.

[AlisdairO]

@GuillaumeGomez Thanks a lot - I've been going through and marking them off in the spreadsheet, but I think we just hit an unfortunate coincidence :-)

[GuillaumeGomez]

On which one ?

[AlisdairO]

On E0139.

[GuillaumeGomez]

What's about E0139 ? It was marked as merged ?

[AlisdairO]

I don't think so - at least, as I see the spreadsheet I'm still marked as the assignee. I see from the issue that I'm supposed to comment in there too, which I haven't been doing, so I'll do that in future!