Add failing regression test for opaque types in #1973

[saleemrashid]

The proposed fix #1984 changes the way unions are emitted, but the issue described in #1973 also affects homogenous aggregates emitted as opaque types. This pull request adds a similar integration test for that case.

[highfive]

Warning

• These commits modify unsafe code. Please review it carefully!

[emilio]

This looks ok, though honestly passing opaque types by value is possibly broken in other cases, and I'm not sure how fixable / easy to fix it'd be, since it's obviously ABI-dependent.

[saleemrashid]

Isn't handling opaque types by value the main reason for opaque types? Otherwise you could use an unsized type and only operate on pointers.

[emilio]

Isn't handling opaque types by value the main reason for opaque types? Otherwise you could use an unsized type and only operate on pointers.

Not really. At least when I implemented it the main use case was "I have this class/struct with this very complex type inside which I don't care about, but I care about the class having the right layout".

Off hand I don't know how we'd go about fixing this (without ABI-specific knowledge), tbf.

[saleemrashid]

Ah, right. I don't really understand when you would require the correct layout for an opaque type other than to pass by value (or to allocate memory/copy memory/etc.).

[emilio]

To access other fields? You can pass a pointer to something and poke at it from rust.

[saleemrashid]

Oh, sorry, I understand now. For when you have a struct with some fields as opaque and some not.

[bors-servo]

☔ The latest upstream changes (presumably 2cfc64e) made this pull request unmergeable. Please resolve the merge conflicts.

[emilio]

Let's close this for now since I don't think we want to codify this behavior. Passing opaque types by value is not guaranteed to work.