Version 0.8.0 checklist about proptest

[AltSysrq]

it also means that you are not permitted to use turbofish at all

Yep, which is why I did it across the board now rather than planning on adopting it opportunistically. IMHO the way it simplifies function signatures in the docs is worth the small risk that it breaks someone's code, especially since we have a whole slew of minor breaking changes in this version anyway.

[AltSysrq]

0.8.0 is released

[AltSysrq]

rustc version 1.27 … 9 days

Not too bad of a delay I guess, and the things it adds seem worth it.

proptest_derive

I'd personally vote for having a minimum viable release first (eg, without filtering/prob/recursive types) and then build up from there incrementally rather than a "big bang" release since it doesn't seem like there's a strong need for them to come all at once.

showable & shrinkable closures

I don't feel like this is worth blocking 0.8.0 on; it's not like adding it will be a huge breaking change.

[Centril]

I'd personally vote for having a minimum viable release first

I think that works; I think (but I'm not 100% sure) that prob and filter will be strictly additive (no breakage) and recursive logic will most likely need to be opt in and will also probably require some annotations from the user because of undecidability in dealing with mutually recursive types in general.

I don't feel like this is worth blocking 0.8.0 on; it's not like adding it will be a huge breaking change.

Agreed -- I'm quite sure that it is a purely additive change;
Hopefully, I'll be fast enough to land this before 1.27 lands anyways ;)

[AltSysrq]

Since 1.27 only stabilises the arch::{x86,x86_64} modules and not simd, is it really worth adding support for that now? The types there are fairly low-level and weakly-typed, which makes precludes any one-size-fits-all support.

[Centril]

Not sure; it should still be possible to implement __m256 and such through _mm256_set_ps. Should be fairly easy to get good coverage with macros?

My primary motivation for 1.27 is RangeInclusive tho and to make API consistent on that front.

There are some FIXMEs and TODOs that are unrelated to 1.27 that I'd like you to look at... primarily in https://github.com/AltSysrq/proptest/blob/0.8.0-changes/src/bits.rs#L30-L31 and https://github.com/AltSysrq/proptest/blob/0.8.0-changes/src/sample.rs#L55-L56 since I am less familiar with those parts..

(Probably won't have time to finish proptest_derive so that it is polished for a release unfortunately if you want 0.8 shipped in the coming week... but I don't think proptest_derive influences decisions about proptest 0.8 -- other than to have a nicer release notes)

(Haven't had time to work enough on Showable and Shrinkable Closures -- so it should for sure not block 0.8)

[AltSysrq]

it should still be possible to implement __m256 and such

Possible, yes, but the types aren't well-defined enough for a one-size-fits-all solution.

some FIXMEs and TODOs that are unrelated to 1.27 that I'd like you to look at

Addressed those and all others that looked immediately fixable. Let me know if I missed something; if not, I'll get 0.8.0 published fairly soon.

[Centril]

Possible, yes, but the types aren't well-defined enough for a one-size-fits-all solution.

Alright; Given that we bumped to 1.27, it can be easily added without breakage if we find a good way in the future and someone needs implementations of those :)

Addressed those and all others that looked immediately fixable. Let me know if I missed something; if not, I'll get 0.8.0 published fairly soon.

Nice work! ❤️

I have one concern re. the movement towards impl Trait in argument position... While it is nice, it also means that you are not permitted to use turbofish at all (not even to specify those type variables which are explicitly quantified that remain). The instances you changed don't seem likely to ever need turbofish, but you never know. I primarily want to highlight the trade-off. I don't mind it being changed to arg: impl Trait :)

Let me know if I missed something;

https://blog.rust-lang.org/2018/06/21/Rust-1.27.html

In Rust 1.27 you can use dyn Trait, which we should do instead of the bare trait syntax where trait objects are currently involved. That's just a matter of style tho and can be fixed in a patch version, so it is not urgent.

There's also #[must_use] which can now be put on functions, which could help users in some cases since strategies usually do nothing unless they are used, for example. In some cases, #[must_use] should probably be placed on the Strategy types themselves. For example, Iterators in the standard library typically say:

#[must_use = "iterator adaptors are lazy and do nothing unless consumed"]

https://doc.rust-lang.org/nightly/std/iter/struct.Map.html

[AltSysrq]

dyn Trait

Right, I had intended on moving to that. Thanks for pointing it out.

#[must_use]

Good suggestion, I'll see where that applies.

[Centril]

Hurray! 🦀