A "Design FAQ" document

[bstrie]

In the vein of the Project FAQ and the Language FAQ.

In addition to being really interesting, this would save a lot of time and divert a lot of potential questions in the long run. Some topics to discuss:

• Why was typestate removed?
• Why are some types nominal instead of structural?
• Why do I have to declare impls in the same crate as a type or trait?
• Why is trait inheritance the only means of inheritance?
• Why doesn't Rust do tail-call optimization?
  (see e.g. rust-dev post.)

Explanations should be technical and include code examples wherever possible.

Some of the questions of this nature overlap with the latter half of the Language FAQ, but not all, e.g. "Can Rust code call C code?" and "What is the difference between a managed pointer and an owned pointer?" do not belong. The document should focus on large-scale design decisions and tradeoffs; think "Why split stacks?" rather than "How do split stacks work?".

If you think of good questions, add them in the comments here. There are probably a lot of ripe areas for explanation, especially with regard to mutability and regions.

[brson]

The manual has a related (and empty) section: http://dl.rust-lang.org/doc/0.4/rust.html#appendix-rationales-and-design-tradeoffs

[bstrie]

Why are &str, @str, ~str, &[], @[], and ~[] treated differently from &T, @T, and ~T?

[bstrie]

Why are early returns disallowed within closures?

[bstrie]

Why are separate do and for constructs required?

[bstrie]

Why do traits in Rust depart from the design in the original traits paper?

[bstrie]

Why must imports be qualified relative to the crate?

[graydon]

These are great, please keep them coming

[catamorphism]

"What does const mean (and why)?" should be in there too, though I know we're thinking about renaming it (at least).

[bstrie]

Why feature conditions rather than exceptions?

[jruderman]

Why so many pointer types? Why not rely on escape analysis instead?

Answered at:

http://www.reddit.com/r/programming/comments/17hqg4/a_little_bit_rusty_practical_approach_on_rust/c85r1nc

[bstrie]

Why is the .times() method no longer defined on signed integers? (discussion on the specifics of integer literal inference)

[bstrie]

Why do borrow checks on @mut pointers happen at runtime?

[acertain]

Why are there no exceptions?

[bstrie]

Why is name resolution such an intractably hard problem?

<@pcwalton> I know of no other static language with declarative imports
            that allows (a) reexports; (b) glob import; (c) mutual
            recursion everywhere; (d) nested modules; (e) two namespaces
<@pcwalton> java doesn't have reexports, Go doesn't have nested modules
            or reexports, C# doesn't have reexports I don't think, C++
            doesn't have reexports or mutual recursion everywhere, maybe
            D...
<@pcwalton> ML and Haskell don't have mutual recursion everywhere and
            don't have reexports
<@pcwalton> mutually recursive modules too
<@pcwalton> most languages do not have mutually recursive modules
< bstrie> pcwalton: so what you're saying is that rust has once again
          failed to be unoriginal :)
<@pcwalton> well, as you said, Python
<@pcwalton> but Rust is trying to achieve Python's module system's
            expressive power in a static system
  jld tries to remember how the various Scheme module systems work
<@pcwalton> turns out that's pretty hard :)
<@pcwalton> jld: no mutual recursion
<@pcwalton> btw, ES6 hit the exact same problems we did
<@pcwalton> and had to pare down their module system as a result
<@pcwalton> anyway none of this is an excuse, we should make it easier to
            use
<@pcwalton> but I'm just saying it's not that easy...

[brson]

Here's a summary about dynamically sized types https://mail.mozilla.org/pipermail/rust-dev/2013-April/003933.html

[brson]

Summary of 'why no exceptions' https://mail.mozilla.org/pipermail/rust-dev/2013-April/003815.html

[catamorphism]

Summary of 'why no purity-in-the-type-system, i.e., why no effect system': https://mail.mozilla.org/pipermail/rust-dev/2013-April/003926.html

[iampiyush]

document how to make non-utf8 byte vectors with bytes!(…)

e.g.
use of bytes!("hell", 0xc0, 0xca, "o!") instead of bytes!("hell\xc0\xca\o1!")

[brson]

-> in function syntax: https://mail.mozilla.org/pipermail/rust-dev/2013-July/005042.html

[kud1ing]

Why have we mostly moved from internal iterators to external iterators?
https://mail.mozilla.org/pipermail/rust-dev/2013-June/004364.html

[kud1ing]

Why was the runtime ported from C++ to Rust?

[brson]

Why is rustc slow? https://mail.mozilla.org/pipermail/rust-dev/2013-November/006975.html

[pnkfelix]

Why does one need to explicitly pull a trait into scope in order to access that trait's methods on a (separately imported or locally defined) type?

Answer (I think): Because the alternative is name collisions (that could arise much later in the dev cycle). We're just forcing you to resolve those conflicts up front. See also: the wiki

[brson]

Why does Rust have two threading implementations?

Why does Rust prefer asynchronous channels?

Why are struct fields public by default?

[brson]

Why are ints pointer sized?

[brson]

Why such a verbose syntax for mutable locals? (Why not var instead of let mut?)

Some links:

https://mail.mozilla.org/pipermail/rust-dev/2014-January/008364.html
https://mail.mozilla.org/pipermail/rust-dev/2014-January/008319.html
#2643

[brson]

Why no TCO (this comes up all the time): https://mail.mozilla.org/pipermail/rust-dev/2013-April/003557.html

[brson]

Why do Rust's submodules map to directories and mod.rs the way they do instead of [some other way]?

https://mail.mozilla.org/pipermail/rust-dev/2014-April/009554.html

[bstrie]

Why a dedicated loop keyword?

[brson]

Why no mutable fields? http://www.reddit.com/r/rust/comments/264d2t/history_question_why_was_mut_from_structsclasses/chnkty5

[kud1ing]

See also #14370

[bstrie]

Why do numeric types wrap on overflow? See also http://blog.regehr.org/archives/1154

(not so relevant anymore. -nmatsakis)

[bstrie]

Why must mod follow use?

(not so relevant anymore. -nmatsakis)

[steveklabnik]

Now that #14370 has been closed, can we close this?

[alexcrichton]

This thread still has many questions which aren't present in the FAQ, so we may in theory want to add all of them to the FAQ before closing this.

[brson]

Agree. We've collected a lot of info here that I don't want to lose.

[pnkfelix]

"Why does let _ = ... behave differently from let _x = ..." : #15151 , #10488 (comment)

[alexchandel]

"Why is there no float type?" or "Why is there int and uint but not float?"

[theemathas]

Why do we need Copy if we already have Clone?

[theemathas]

Why is moving function arguments the default, not borrowing or copying/cloning?

[alexchandel]

"Why is i32 the default?" and "Why aren't isize and usize named after C's intptr_t and uintptr_t?"

[steveklabnik]

I'm pulling a massive triage effort to get us ready for 1.0. As part of this, I'm moving stuff that's wishlist-like to the RFCs repo, as that's where major new things should get discussed/prioritized.

This issue has been moved to the RFCs repo: rust-lang/rfcs#615

[pnkfelix]

Oh no! It was really nice that this list was all in one place!

[pnkfelix]

2015Feb20 10:23 <eddyb> wrote:

pnkfelix: do we have a FAQ-style answer for "why disallow destructuring types that impl Drop"?

[steveklabnik]

I'm pulling a massive triage effort to get us ready for 1.0. As part of this, I'm moving stuff that's wishlist-like to the RFCs repo, as that's where major new things should get discussed/prioritized.

This issue has been moved to the RFCs repo: rust-lang/rfcs#891

Update from pnkfelix: No, that attempted move failed. See rust-lang/rfcs#615

[alexchandel]

I guess closing all your issues is a nice way to pretend they're not issues. Adding an FAQ shouldn't need to go through the RFC process.

[steveklabnik]

@alexchandel the RFC repo issues are not just for things going through the process. It's also full of suggestions for things like community libraries that could be done. The result of this won't need to go through the RFC process, it's just that I don't plan on writing this any time soon.

[pnkfelix]

Note also that there are prototype FAQ documents already in src/doc:

• https://github.com/rust-lang/rust/blob/master/src/doc/complement-design-faq.md
• https://github.com/rust-lang/rust/blob/master/src/doc/complement-lang-faq.md
• https://github.com/rust-lang/rust/blob/master/src/doc/complement-project-faq.md

(Nonetheless, these tickets remain worthwhile catch-all lists of questions that may not yet have answers on any of those three documents. I'm still not thrilled with the pseudo-migration to the RFC repo.)

[steveklabnik]

Yes, I'm not exactly psyched about those ones, either, to be honest.

[pnkfelix]

Why can't I do destructuring-bind in a trait default method?

For an example, see this playpen

See discussion from 2013-09-10, though it seems like the "decision" there was never implemented (and likely never will at this point?)