abi: unsized field in union - assert to delay bug

Unions cannot have unsized fields, and as such, layout computation for
unions asserts that each union field is sized (as this would normally
have halted compilation earlier).

However, if a generator ends up with an unsized local - a circumstance
in which an error will always have been emitted earlier, for example, if
attempting to dereference a `&str` - then the generator transform will
produce a union with an unsized field.

Since #110107, later passes will be run, such as constant propagation,
and can attempt layout computation on the generator, which will result
in layout computation of `str` in the context of it being a field of a
union - and so the aforementioned assertion would cause an ICE.

It didn't seem appropriate to try and detect this case in the MIR body
and skip this specific pass; tainting the MIR body or delaying a bug
from the generator transform (or elsewhere) wouldn't prevent this either
(as neither would prevent the later pass from running); and tainting when
the deref of `&str` is reported, if that's possible, would unnecessarily
prevent potential other errors from being reported later in compilation,
and is very tailored to this specific case of getting a unsized type in
a generator.

Given that this circumstance can only happen when an error should have
already been reported, the correct fix appears to be just changing the
assert to a delayed bug. This will still assert if there is some
circumstance where this occurs and no error has been reported, but it
won't crash the compiler in this instance.

Signed-off-by: David Wood <[email protected]>

Author: davidtwco

compiler/rustc_abi/src/layout.rs

@@ -763,7 +763,9 @@ pub trait LayoutCalculator {
         let mut size = Size::ZERO;
         let only_variant = &variants[FIRST_VARIANT];
         for field in only_variant {
-            assert!(field.0.is_sized());
+            if field.0.is_unsized() {
+                self.delay_bug("unsized field in union".to_string());
+            }
 
             align = align.max(field.align());
             max_repr_align = max_repr_align.max(field.max_repr_align());

compiler/rustc_abi/src/lib.rs

@@ -1345,7 +1345,6 @@ impl Abi {
 
     /// Discard validity range information and allow undef.
     pub fn to_union(&self) -> Self {
-        assert!(self.is_sized());
         match *self {
             Abi::Scalar(s) => Abi::Scalar(s.to_union()),
             Abi::ScalarPair(s1, s2) => Abi::ScalarPair(s1.to_union(), s2.to_union()),

tests/ui/generator/issue-113279.rs

@@ -0,0 +1,27 @@
+#![feature(generators)]
+
+// `foo` attempts to dereference `""`, which results in an error being reported. Later, the
+// generator transform for `foo` then produces a union which contains a `str` type - unions should
+// not contain unsized types, but this is okay because an error has been reported already.
+// When const propagation happens later in compilation, it attempts to compute the layout of the
+// generator (as part of checking whether something can be const propagated) and in turn attempts
+// to compute the layout of `str` in the context of a union - where this caused an ICE. This test
+// makes sure that doesn't happen again.
+
+fn foo() {
+    let _y = static || {
+        let x = &mut 0;
+        *{
+            yield;
+            x
+        } += match { *"" }.len() {
+            //~^ ERROR cannot move a value of type `str` [E0161]
+            //~^^ ERROR cannot move out of a shared reference [E0507]
+            _ => 0,
+        };
+    };
+}
+
+fn main() {
+    foo()
+}

tests/ui/generator/issue-113279.stderr

@@ -0,0 +1,16 @@
+error[E0161]: cannot move a value of type `str`
+  --> $DIR/issue-113279.rs:17:20
+   |
+LL |         } += match { *"" }.len() {
+   |                    ^^^^^^^ the size of `str` cannot be statically determined
+
+error[E0507]: cannot move out of a shared reference
+  --> $DIR/issue-113279.rs:17:22
+   |
+LL |         } += match { *"" }.len() {
+   |                      ^^^ move occurs because value has type `str`, which does not implement the `Copy` trait
+
+error: aborting due to 2 previous errors
+
+Some errors have detailed explanations: E0161, E0507.
+For more information about an error, try `rustc --explain E0161`.