Auto merge of #115524 - RalfJung:misalign, r=wesleywiser

const-eval: make misalignment a hard error

It's been a future-incompat error (showing up in cargo's reports) since rust-lang/rust#104616, Rust 1.68, released in March.  That should be long enough.

The question for the lang team is simply -- should we move ahead with this, making const-eval alignment failures a hard error? (It turns out some of them accidentally already were hard errors since #104616. But not all so this is still a breaking change. Crater found no regression.)

src/machine.rs

@@ -12,7 +12,6 @@ use rand::rngs::StdRng;
 use rand::SeedableRng;
 
 use rustc_ast::ast::Mutability;
-use rustc_const_eval::const_eval::CheckAlignment;
 use rustc_data_structures::fx::{FxHashMap, FxHashSet};
 #[allow(unused)]
 use rustc_data_structures::static_assert_size;
@@ -885,28 +884,15 @@ impl<'mir, 'tcx> Machine<'mir, 'tcx> for MiriMachine<'mir, 'tcx> {
     const PANIC_ON_ALLOC_FAIL: bool = false;
 
     #[inline(always)]
-    fn enforce_alignment(ecx: &MiriInterpCx<'mir, 'tcx>) -> CheckAlignment {
-        if ecx.machine.check_alignment == AlignmentCheck::None {
-            CheckAlignment::No
-        } else {
-            CheckAlignment::Error
-        }
+    fn enforce_alignment(ecx: &MiriInterpCx<'mir, 'tcx>) -> bool {
+        ecx.machine.check_alignment != AlignmentCheck::None
     }
 
     #[inline(always)]
     fn use_addr_for_alignment_check(ecx: &MiriInterpCx<'mir, 'tcx>) -> bool {
         ecx.machine.check_alignment == AlignmentCheck::Int
     }
 
-    fn alignment_check_failed(
-        _ecx: &InterpCx<'mir, 'tcx, Self>,
-        has: Align,
-        required: Align,
-        _check: CheckAlignment,
-    ) -> InterpResult<'tcx, ()> {
-        throw_ub!(AlignmentCheckFailed { has, required })
-    }
-
     #[inline(always)]
     fn enforce_validity(ecx: &MiriInterpCx<'mir, 'tcx>, _layout: TyAndLayout<'tcx>) -> bool {
         ecx.machine.validate