Rollup merge of rust-lang#58615 - RalfJung:ref-to-place-alignment, r=…

…oli-obk

miri: explain why we use static alignment in ref-to-place conversion

@eddyb @oli-obk do you think this makes sense? Or should we use the run-time alignment (`align_of_val`)? I am a bit worried about custom DSTs, but that affects way more areas of Miri so I'd ignore them for now.

r? @oli-obk

src/librustc_mir/interpret/place.rs

@@ -326,6 +326,10 @@ where
 
         let mplace = MemPlace {
             ptr: val.to_scalar_ptr()?,
+            // We could use the run-time alignment here. For now, we do not, because
+            // the point of tracking the alignment here is to make sure that the *static*
+            // alignment information emitted with the loads is correct. The run-time
+            // alignment can only be more restrictive.
             align: layout.align.abi,
             meta: val.to_meta()?,
         };
@@ -385,9 +389,11 @@ where
         // above). In that case, all fields are equal.
         let field_layout = base.layout.field(self, usize::try_from(field).unwrap_or(0))?;
 
-        // Offset may need adjustment for unsized fields
+        // Offset may need adjustment for unsized fields.
         let (meta, offset) = if field_layout.is_unsized() {
-            // re-use parent metadata to determine dynamic field layout
+            // Re-use parent metadata to determine dynamic field layout.
+            // With custom DSTS, this *will* execute user-defined code, but the same
+            // happens at run-time so that's okay.
             let align = match self.size_and_align_of(base.meta, field_layout)? {
                 Some((_, align)) => align,
                 None if offset == Size::ZERO =>