derive: assume enum repr defaults to isize

[durka]

derive: assume enum repr defaults to isize

Fixes#31886.

Spawned from #32139.

r? @alexcrichton

[alexcrichton]

@bors: r+ e45b59aed463c15bd9aadd8280f4a5d49027e73f

[bors]

☔ The latest upstream changes (presumably #30587) made this pull request unmergeable. Please resolve the merge conflicts.

[durka]

@oli-obk is this fix now irrelevant given your PR?

[oli-obk]

well kindof.

I actually broke the following construct (explicitly, not accidentally) on 32bit

enumEu64{Au64 = 0,Bu64 = 0x8000_0000_0000_0000}

because on 32bit rustc now errors at compiletime about the collision

Once we properly fix the automatic enum discriminant type detection, the issue might appear again.

[durka]

@oli-obk can you clarify what it means to properly fix discriminant detection, and what the current situation is? Is deriving correct in assuming that if the repr isn't specified than it will be isize (or smaller)?

[oli-obk]

@durka

In list form, because I'm confusing myself when I'm writing it in text form.

1. -9223372036854775808isize as u64 always was 0 on 32bit and still is.
2. Eu64::Bu64 as u64 always was 0 on 32bit and still is
3. 2. makes zero sense, as the enum variant discriminants can't both be zero
4. size_of::<Eu64>() == 8 on 32bit and 64bit
5. I broke the mentioned enum, because my typestrong const int branch ends up turning the discriminant of Eu64::Bu64 to 0 early enough to allow rustc to detect the collision.
6. Now there's no way to specify a single value that ends up turning the discriminant to be 8 bytes. (the Ei64 example uses two values, a negative one and a large positive one)
7. I'm not sure how to reproduce very slow derived PartialEq/PartialOrd implementation of enum #31886 now, but Ei64 is still 8 bytes, while the discriminant type is isize.

can you clarify what it means to properly fix discriminant detection

well... the discriminant should be i64 and u64 imo, it should never assume isize/usize automatically, because that's really not portable at all. If someone wants pointer-size, then they should annotate it with repr. But that would technically be a breaking change for anyone ignoring the literal overflow warnings.

[durka]

@oli-obk the code given in #31886 still builds (and hangs) on nightly-2016-03-16.

[oli-obk]

oh.... I only tested on 32 bit... it obviously only collides on 32 bit... So yea, this change is still necessary for 64 bit.

[durka]

Okay, so I changed it back to isize since now @oli-obk's changes should prevent an enum with colliding variants from compiling.

Note that you can still write a program like this:

enumEu64{Au64 = 0,Bu64 = 0x8000_0000_0000_0001//~WARN literal out of range for isize}fnmain(){println!("{} {} {}", std::mem::size_of::<Eu64>(),Eu64::Au64asisize,Eu64::Bu64asisize);}

which prints different answers on 32- and 64-bit (1 0 1 and 8 0 -9223372036854775807 respectively).

[durka]

And combined with the fact that #[derive(PartialEq, PartialOrd)] uses discriminant_value, the above enum will give different answers for Au64 < Bu64...

I guess that's the price you pay for ignoring the warnings.

[alexcrichton]

Sorry I've gotten a bit lost with all the changes in flight. Is this ready to go? Did something about constants change how we want to do this?

[durka]

@alexcrichton what I believe is that is no longer possible to trigger #31886 on a 32-bit system because @oli-obk's const evaluator will catch the collision. However, my fix (as implemented currently, in this PR!) is still needed to avoid #31886 on 64-bit systems.

[alexcrichton]

@bors: r+ 9d0748f

Ok, thanks for the explanation @durka!

[bors]

⌛ Testing commit 9d0748f with merge 2ddcf31...

[bors]

💔 Test failed - auto-win-msvc-64-opt-rustbuild

[oli-obk]

spurious failure

Makefile:26: recipe for target 'clean' failed
Zombie process: "\Device\HarddiskVolume1\dojob.exe"
thread '

' panicked at 'fs::remove_dir_all(path) failed with Access is denied. (os error 5)', C:\bot\slave\auto-win-msvc-64-opt-rustbuild\build\src\bootstrap\build\clean.rs:34
note: Run with RUST_BACKTRACE=1 for a backtrace.
make: *** [clean] Error 101

[alexcrichton]

@bors: retry

[bors]

⌛ Testing commit 9d0748f with merge 6cc502c...

[bors]

☀️ Test successful - auto-linux-32-nopt-t, auto-linux-32-opt, auto-linux-64-debug-opt, auto-linux-64-nopt-t, auto-linux-64-opt, auto-linux-64-opt-rustbuild, auto-linux-64-x-android-t, auto-linux-64-x-armhf, auto-linux-64-x-armsf, auto-linux-64-x-freebsd, auto-linux-64-x-netbsd, auto-linux-cross-opt, auto-linux-musl-64-opt, auto-mac-32-opt, auto-mac-64-nopt-t, auto-mac-64-opt, auto-mac-ios-opt, auto-win-gnu-32-nopt-t, auto-win-gnu-32-opt, auto-win-gnu-32-opt-rustbuild, auto-win-gnu-64-nopt-t, auto-win-gnu-64-opt, auto-win-msvc-32-opt, auto-win-msvc-64-opt, auto-win-msvc-64-opt-rustbuild