Compile with stable Rust

[SergioBenitez]

The following features need to be stabilized before Rocket can compile on stable:

• [x] proc_macro_hygiene (rust-lang/rust#54727)
• [x] associated_consts (rust-lang/rust#29646)
• [x] conservative_impl_trait (rust-lang/rust#34511)
• [x] drop_types_in_const (rust-lang/rust#33156)
• [x] more_io_inner_methods (rust-lang/rust#41519)
• [x] pub_restricted (rust-lang/rust#32409)
• [x] struct_field_attributes (rust-lang/rust#41681)
• [x] use_extern_macros (rust-lang/rust#35896)
• [x] fn_box (rust-lang/rust#28796)
• [x] try_from (rust-lang/rust#33417)

The following dependencies rely on nightly features and must be updated before Rocket can compile on stable:

• [x] pear (https://github.com/SergioBenitez/Pear/commit/333c9bf5e2c516c23b17c7482b410f1448c1b3af)

Update (Feb 07, 2017): Added list of features used by Rocket. Update (Feb 28, 2017): Added lookup_host feature. Update (Mar 21, 2017): pub_restricted was stabilized! Update (Mar 30, 2017): Added never_type feature. Update (Apr 16, 2017): Added concat_idents feature. Update (May 19, 2017): Added struct_field_attributes and more_io_inner_methods features. Update (May 19, 2017): concat_idents is no longer used. Update (May 19, 2017): Added links to tracking issues. Update (May 20, 2017): type_ascription is no longer used. Update (Jun 19, 2017): struct_field_attributes was stabilized! Update (Jun 24, 2017): Added try_trait feature. Update (Jul 1, 2017): more_io_inner_methods was stabilized! Update (Jul 9, 2017): associated_consts was stabilized! Update (Sep 7, 2017): lookup_host is no longer used. Update (Sep 14, 2017): drop_types_in_const was stabilized! Update (Sep 14, 2017): Added decl_macro feature. Update (Mar 26, 2018): conservative_impl_trait, never_type, and i128_type were stabilized! Update (Apr 22, 2018): Added fnbox feature. Update (Apr 26, 2018): never_type stabilization was reverted (rust-lang/rust#50121). Update (May 5, 2018): Swapped macro_reexport for use_extern_macros. Update (Jul 29, 2018): Added crate_visibility_modifier and try_from features. Update (Aug 5, 2018): custom_derive is no longer used. Update (Aug 17, 2018): use_extern_macros was stabilized! Update (Sep 3, 2018): const_fn is no longer used. Update (Sep 26, 2018): Added label_break_value feature. Update (Oct 9, 2018): Removed compiler plugin features (custom_attribute, plugin). Update (Oct 9, 2018): Updated proc_macro features. Update (Mar 9, 2019): try_from was stabilized! Update (Apr 13, 2019): fnbox is no longer used. Update (Jul 9, 2019): never_type is no longer used. Update (Jul 19, 2019): decl_macro is no longer used. Update (Sep 9, 2019): specialization is no longer used. Update (Sep 10, 2019): label_break_value is no longer used. Update (Sep 18, 2019): try_trait is no longer used. Update (Sep 21, 2019): crate_visibility_modifier is no longer used. Update (May 19, 2020): proc_macro_hygiene was stabilized! Update (Jul 16, 2020): proc_macro_diagnostics is no longer used. Update (Jul 16, 2020): proc_macro_span is no longer used. Update (Jul 21, 2020): pear was updated to be stable-compatible.

[steveklabnik]

Looks like this primarily depends on https://github.com/rust-lang/rust/issues/38356

(Or syntex)

[rrichardson]

Syntex should get this onto stable much more quickly, since the macros > 1.1 effort seems to be just beginning to congeal.

https://news.ycombinator.com/item?id=13245975

[steveklabnik]

since the macros > 1.1 effort seems to be just beginning to congeal.

This needs more than macros 1.1, it needs 2.0. (If it only needed 1.1, then it could go stable in 1.15)

[rrichardson]

I intended > 1.1 to indicate "after 1.1" but that wasn't very clear.
I've seen the procedural macros effect referenced as 1.2 as well as 2.0. 2.0 does make more sense, since it is a much broader scope, as I understand it.

[mgattozzi]

I've started working on getting syntex to work with this until macros are stabilized (which might be a while). I've got time over the holidays from work so I'll post any progress I make!

[SergioBenitez]

Rocket uses much more than just plugins from nightly. The set of features Rocket currently uses are:

#![feature(plugin)]
#![feature(custom_derive)]
#![feature(custom_attribute)]
#![feature(specialization)]
#![feature(conservative_impl_trait)]
#![feature(drop_types_in_const)]
#![feature(associated_consts)]
#![feature(const_fn)]

These uses are important and contribute to the pithiness of the API, so they will stick around until they're stabilized. Until then, I don't believe Rocket will run on stable.

With that being said, I believe that the ergonomics of syntex are very poor, and I don't want to recommend a subpar solution to users. I'm all for getting Rocket to work on stable as soon as possible, but I'm not willing to compromise ergonomics and usability to make that happen.

[beefsack]

I'm all for getting Rocket to work on stable as soon as possible, but I'm not willing to compromise ergonomics and usability to make that happen.

I'd love to see it go stable as soon as possible, but I feel you've got the right values for the project. Accessibility is arguably more important for web frameworks than it is for other frameworks and libraries.

[mgattozzi]

Just from trying to get it to work with syntex it's hard. I might try some more this week but you're right. It might just be worth it to wait till those features stabilize.

[tbu-]

The lookup_host feature can be replaced by the ToSocketAddrs implementation of &str.

[FlogramMatt]

Perhaps you could use something like: https://github.com/Kimundi/rustc-version-rs

In order to allow stable features to compile using stable, but features that require nightly or beta could be used only when compiling from that branch.

Currently using Iron myself and would love to switch but it's a high security project and for security reasons, I want to avoid Rust nightly.

[FlogramMatt]

It's been a year since this post originally... seems the check boxes in the original post are accurate?

Seems to me like getting Rocket out of nightly would be a huge boost for Rust in general but especially for Rocket.. the Rocket team or Rust team should really focus on stabilizing these functions (or finding alternatives?).

Why so slow to get there? Have you run this by the Rust higher ups and pushed them a little bit?

[messense]

never_type will be stabilized in Rust 1.26.0.

[messense]

impl Trait is scheduled to be stabilized in Rust 1.26.0 too

[est31]

i128 just got stabilized will ship in Rust 1.27.0.

[tsal]

I'm really looking forward to using Rocket in production some day, I'm glad to see there's been some progress recently. :)

[messense]

macro_reexport is now removed(not in nightly yet, incoming breaking change) and subsumed by feature(use_extern_macros) and pub use

https://github.com/rust-lang/rust/pull/49982

[maxdeviant]

@messense PR #624 should handle those changes.

[mentalbrew]

The Rust community is currently in a major push toward releasing the Rust 2018 Edition. Has anybody investigated whether enough features that Rocket depends on in nightly will be available in that edition that the Rocket community could target that release for stable? It seems like it would be a win-win for both efforts if it was remotely feasible.

[jebrosen]

I glanced at the 2018 feature status and most of the important ones for Rocket are not listed:

• const_fn
• decl_macro
• never_type
• proc_macro, and several related features such as proc_macro_non_items
• specialization, which can apparently be worked around, but might be problematic to drop for existing rocket users

That does not necessarily mean that these features will not be stabilized this year, just that they are not being targeted as part of the Edition. async/await, however, is listed, and says 2018 will be the minimum edition. Depending on what async solution is implemented, Rocket might target the 2018 edition to make use of those ergonomics.

That being said, editions are cross-crate compatible. Whether Rocket ends up targeting the 2015 or 2018 edition should not significantly affect projects that use it.

[est31]

Just noting, proc_macro is in the process of being stabilized: https://github.com/rust-lang/rust/pull/52081

[SergioBenitez]

@est31 Unfortunately, the stabilizing subset is extremely limited and insufficient for Rocket's needs.

[cauebs]

Now there's an easier way to keep track of these things: https://forge.rust-lang.org/state-of-rust.html

[bestouff]

Either way it'd be nice to have a "tentative ETA" for stable Rocket. Even if it's "don't expect less than 2 years" that could help a lot people who might want to select it but wonder about how long they have to suffer in nightly.

[FlogramMatt]

@SergioBenitez You should really reach out to the Rust team and ask for them to prioritize these issues, tell them it's holding up release of Rocket. Or maybe someone can volunteer to help fix these issues if they'll help point you in the right direction?

I know that getting Rocket-like web servers was being listed as a high priority for getting Rust adoption and Rocket looks like an otherwise amazing one.

Team can be found here: https://www.rust-lang.org/en-US/team.html

Have communicated with Brian in the past who seemed particularly good at responding and helping us out on a far less important issue: https://github.com/brson

[0xcaff]

FYI, if you don't want to wait to build web services in rust, check out actix. It works on stable rust.

[dataf3l]

@0xcaff After trying multiple web frameworks, I have to say really like Rocket. Iron, unfortunately, didn't do it for me...

I'm grateful for the efforts made by the iron team, and I value open source contributor's time as much as the next guy, but for me, I think Rocket suits fits my needs better.

[ErichDonGubler]

proc_macro has been split into several subtasks -- which one(s) does Rocket depend on?

EDIT: Thanks!

[Xaeroxe]

The following workaround has been used by the Amethyst game engine for the absence of a never type

pub enum Never {}

This type can't be instantiated and thus fills the same purpose.

[BanjoFox]

Very happy to see this progressing! Team Aardwolf appreciates your hard work :D

[jhpratt]

Given the recent blogpost, shouldn't the milestone on this be 0.5?

[est31]

@incubus8 see this blog post for details on the plans: https://rocket.rs/v0.4/news/2018-12-08-version-0.4/

[rhzs]

gotcha, sorry for the fuzz

@incubus8 see this blog post for details on the plans: https://rocket.rs/v0.4/news/2018-12-08-version-0.4/

[macpp]

Just noting: on the latest nightly uniform_paths were stabilised, together with ability to export macro_rules from modules:

mod foo {
    macro_rules! bar123 { () => {} }
    pub(crate) use bar123 as bar;
}
foo::bar!();

It seems that it may be possible to remove decl_macro feature ( when targeting 2018 edition)

[jebrosen]

It seems that it may be possible to remove decl_macro feature ( when targeting 2018 edition)

I played with the new export functionality you mention; it's unfortunate that it only works when the application crate targets 2018. I think it would be acceptable to implement the change if 2018 is required for another reason (e.g. async/await). But I wouldn't want to break compatibility with 2015 edition just to avoid decl_macro unless it's the only feature blocking stable, and maybe not even then.

[Centril]

@jebrosen I would suggest avoiding long term reliance on decl_macro; I don't think the language team has any bandwidth to devote the substantial time that decl_macro requires to stabilize it.

[abonander]

std::convert::Infallible is stable in the upcoming release, that should be able to substitute for most uses of the never-type.

[jhpratt]

^^ I can quickly go through and issue a PR if that's something that's wanted.

[jebrosen]

At the very least, a PR will make it obvious where it's currently used and whether Infallible can fully replace its uses. It may not be accepted right way, especially if it would break existing code that uses the real never type.

[jhpratt]

@jebrosen Is there a minimum version Rocket guarantees, or is it just the current version? Asking because the initial checkout and cargo check returns warnings for both try_from and transpose_result. I'll naturally remove those as well if desired.

[jebrosen]

@jhpratt Only the Infallible change in one PR (at the very least separate commits) will be easier to split up later if we want to.

[Henning-K]

reqwest is almost stabled It's currently on 0.9.17 https://github.com/seanmonstar/reqwest

And how, if I may ask, is that relevant here?

[sound2gd]

hello, any update on this?

[jhpratt]

@sound2gd Rocket itself should be able to compile on stable as of today's release. Codegen is what still needs nightly.

[mastfissh]

If we wanted to use rocket on stable and we use codegen right now, how hard do you think it would be to remove the codegen usage?

Presumably we could use something like cargo-expand to see the generated code, and then replace the macro invocation with the generated code?

[jebrosen]

IIRC Some of rocket's regular dependencies in master still use nightly features (most/all fixable by updating to a more recent version) and rocket depends on `rocket_codegen. Therefore it would take some work to actually do so. Using Rocket without codegen is usually recommended against and loses most of Rocket's value-add, so it's not something I'm interested in investing time into in favor of working on async and other improvements for 0.5.

Presumably we could use something like cargo-expand to see the generated code, and then replace the macro invocation with the generated code?

I have usually found this to be impractical for anything other than testing, because you wouldn't want to and can't use the expansion of certain macros like format!(). In principle it could probably be done, but would not be very nice to maintain.

[mastfissh]

To summarize - it's probably not useful, easy or very maintainable to use rocket without codegen. Makes sense.

Thanks for the clear explanation 👍

[KamilaBorowska]

never_type will be stabilized 1.41.0, so you can change std::convert::Infallible into !

[nicoburns]

Is there any chance we could use this trick to remove the dependency on proc_macro_hygiene, and get Rocket compiling on stable?

This reddit discussion suggests that it's a lot simpler than proc-macro-hack.

[jebrosen]

I think we could use that trick and get halfway there (we will have some diagnostics-related features to work out), but if I'm reading it right it would limit each function from calling routes!() or uri!() once.

[Kampfkarren]

Why doesn't Rocket just use proc-macro-hack directly? Implementation difficulty?