Debug::fmt vs Display::fmt

matklad · 2018-12-22T09:07:41.259Z

Curious, why Debug & Display have use the same fmt method name? I think it could cause problems with ambiguity if you have both traits in scope (IIRC, something similar happened for backtrace crate recently, whe adding a Display impl broke the clients, most notably failure).

Wouldn’t it be better and more idiomatic to have debug and display methods?

bluss · 2018-12-22T15:10:46.890Z

I disagree that it is necessary to disambiguate them. All the formatting traits share the same method name, so any proposal could take all of them into account. I’ve found that they always disambiguate in my use by which trait bounds are used on the type they are used on.

matklad · 2018-12-22T15:31:30.654Z

I disagree that it is necessary to disambiguate them.

It does happen in practice if you work with concrete types. Here’s an example of a very real ecosystem breakage due to this.

github.com/rust-lang-nursery/failure

Use associated function syntax for `Debug` and `Display` backtrace impl

by azriel91 on 06:42PM - 11 Dec 18 UTC

4 commits changed 4 files with 6 additions and 6 deletions.

in my own code, I tend to use fmt::Debug::fmt as well, because I’ve had similar conflicts when implementing both debug and display for my types.

Soni · 2018-12-22T15:34:16.442Z

Technically you’re not supposed to use those directly, but instead are expected to use write!(f, “{}”, foo) or write!(f, “{:?}”, foo) instead. Both of which should optimize to the respective direct calls, and look more obvious. They also don’t require importing the traits.

bluss · 2018-12-22T15:37:26.988Z

Oh I see, that makes sense. Unfortunate but the resolution to call Debug::fmt(x, f) looks easy to read as well.

In general there is a difference of using fmt versus going through write! and I guess they want to do the former here.

matklad · 2018-12-22T18:17:26.254Z

Yeah, io/fmt Write is a different business: they have to have identical method names for the write! macro to work.

daboross · 2018-12-23T08:01:39.792Z

Doesn’t write!() hard-code ::std::fmt::Debug and ::std::fmt::Dislay for {:?} and {}, though? If it hardcodes those I don’t think it would be an issue hardcoding different method names.

I personally like the names, though, since they’re descriptive and don’t repeat information from the trait.

matklad · 2018-12-24T08:47:16.467Z

I meant that write! macro uses dot call and not UFCS for calling write_fmt on the first argument. This is done to enable “duck typing” and making the same write! macro work with both io::Write and fmt::Write. Calls to Debug/Display indeed can use UFCS.

system · 2019-03-25T08:31:20.635Z

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.