always check overflow in CheckedBinOp in CTFE

compiler/rustc_const_eval/src/interpret/machine.rs

@@ -144,6 +144,9 @@ pub trait Machine<'mir, 'tcx>: Sized {
         true
     }
 
+    /// Whether CheckedBinOp MIR statements should actually check for overflow.
+    fn check_binop_checks_overflow(_ecx: &InterpCx<'mir, 'tcx, Self>) -> bool;
+
     /// Entry point for obtaining the MIR of anything that should get evaluated.
     /// So not just functions and shims, but also const/static initializers, anonymous
     /// constants, ...
@@ -468,6 +471,11 @@ pub macro compile_time_machine(<$mir: lifetime, $tcx: lifetime>) {
         true
     }
 
+    #[inline(always)]
+    fn check_binop_checks_overflow(_ecx: &InterpCx<$mir, $tcx, Self>) -> bool {
+        true
+    }
+
     #[inline(always)]
     fn call_extra_fn(
         _ecx: &mut InterpCx<$mir, $tcx, Self>,

compiler/rustc_const_eval/src/interpret/operator.rs

@@ -32,7 +32,8 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
         );
         // As per https://github.com/rust-lang/rust/pull/98738, we always return `false` in the 2nd
         // component when overflow checking is disabled.
-        let overflowed = overflowed && (force_overflow_checks || self.tcx.sess.overflow_checks());
+        let overflowed =
+            overflowed && (force_overflow_checks || M::check_binop_checks_overflow(self));
         // Write the result to `dest`.
         if let Abi::ScalarPair(..) = dest.layout.abi {
             // We can use the optimized path and avoid `place_field` (which might do

compiler/rustc_middle/src/mir/syntax.rs

@@ -993,8 +993,9 @@ pub enum Rvalue<'tcx> {
 
     /// Same as `BinaryOp`, but yields `(T, bool)` with a `bool` indicating an error condition.
     ///
-    /// When overflow checking is disabled, the error condition is false. Otherwise, the error
-    /// condition is determined as described below.
+    /// When overflow checking is disabled and we are generating run-time code, the error condition
+    /// is false. Otherwise, and always during CTFE, the error condition is determined as described
+    /// below.
     ///
     /// For addition, subtraction, and multiplication on integers the error condition is set when
     /// the infinite precision result would be unequal to the actual result.