repr(i64) picks 32-bit discriminant on 32-bit platform

[bluss]

From the PR #25651 it looks like repr(i64) (and presumably repr(u64)) picks a 32-bit discriminant for the non C-like enum case.

Testcase was:

+#[repr(i64)]
+enum Ei64<T> {
+    _None,
+    _Some(T),
+}
+
+    assert_eq!(size_of::<Ei64<()>>(), 8);

Errored on platform auto-mac-32-opt with this:

thread '<main>' panicked at 'assertion failed:(left == right)(left:4, right:8)', /Users/rustbuild/src/rust-buildbot/slave/auto-mac-32-opt/build/src/test/run-pass/enum-discrim-manual-sizing-2.rs:89

[Aatch]

This is actually pretty simple to fix. Right now we select the type based on, mostly, the overall alignment of the type. We're also basically assuming that an alignment-sized integer is big enough to fit all discriminants in, what happens is that the alignment for a 64-bit integer on a 32-bit platform is 4 bytes, so we end up using a 4-byte integer for discriminant.

The easiest fix is to just skip the calculation for the discriminant type altogether when we have an repr attribute on the enum. A slightly better one might be to only allow the calculation to extend the int type to a larger size.

[Manishearth]

@Aatch could you post a link to the general area of code that needs tweaking?

[devonhollowood]

I'd be happy to work on this if no one else is doing so yet.

[Manishearth]

Go ahead! I'll help if necessary, though @Aatch knows more about this bug.

[Aatch]

The code you need to change is in trans/adt.rs. min_ity is the discriminant type chosen earlier, which will take account of the hint. To do the better fix, you want to add check to see if the alignment type is smaller than the min_ity type and use min_ity instead when setting the actual discriminant type, ity.

[devonhollowood]

I've been having some trouble getting the compiler to build with 32-bit emulation on my 64-bit machine. I thought ./configure --target=i686-apple-darwin or ./configure --host=i686-apple-darwin would work but neither has (it's also possible I'm running the tests wrong--I've been using make check -j4 and make check -j4 TESTNAME=src/test/run-pass/enum-discrim-manual-sizing.rs. As a result I haven't been able to reproduce the bug on my machine. So I submitted a pull request (#32010) containing tests, but no fix, which should hopefully not pass the CI. If the CI rejects it, then it means I have reproduced the bug successfully. Fingers crossed for failure!

[dotdash]

FWIW, I'm using --build=i686-unknown-linux-gnu to cross-compile. I'm doing that seldomly enough not to be sure what exactly is the difference between that, --host and --target.