1 files changed, 64 insertions, 0 deletions
diff --git a/tests/coverage/overflow.rs b/tests/coverage/overflow.rs
new file mode 100644
index 000000000..1c40771b2
--- /dev/null
+++ b/tests/coverage/overflow.rs
@@ -0,0 +1,64 @@
+#![allow(unused_assignments)]
+// compile-flags: -Coverflow-checks=yes
+// failure-status: 101
+
+fn might_overflow(to_add: u32) -> u32 {
+    if to_add > 5 {
+        println!("this will probably overflow");
+    }
+    let add_to = u32::MAX - 5;
+    println!("does {} + {} overflow?", add_to, to_add);
+    let result = to_add + add_to;
+    println!("continuing after overflow check");
+    result
+}
+
+fn main() -> Result<(), u8> {
+    let mut countdown = 10;
+    while countdown > 0 {
+        if countdown == 1 {
+            let result = might_overflow(10);
+            println!("Result: {}", result);
+        } else if countdown < 5 {
+            let result = might_overflow(1);
+            println!("Result: {}", result);
+        }
+        countdown -= 1;
+    }
+    Ok(())
+}
+
+// Notes:
+//   1. Compare this program and its coverage results to those of the very similar test `assert.rs`,
+//      and similar tests `panic_unwind.rs`, abort.rs` and `try_error_result.rs`.
+//   2. This test confirms the coverage generated when a program passes or fails a
+//      compiler-generated `TerminatorKind::Assert` (based on an overflow check, in this case).
+//   3. Similar to how the coverage instrumentation handles `TerminatorKind::Call`,
+//      compiler-generated assertion failures are assumed to be a symptom of a program bug, not
+//      expected behavior. To simplify the coverage graphs and keep instrumented programs as
+//      small and fast as possible, `Assert` terminators are assumed to always succeed, and
+//      therefore are considered "non-branching" terminators. So, an `Assert` terminator does not
+//      get its own coverage counter.
+//   4. After an unhandled panic or failed Assert, coverage results may not always be intuitive.
+//      In this test, the final count for the statements after the `if` block in `might_overflow()`
+//      is 4, even though the lines after `to_add + add_to` were executed only 3 times. Depending
+//      on the MIR graph and the structure of the code, this count could have been 3 (which might
+//      have been valid for the overflowed add `+`, but should have been 4 for the lines before
+//      the overflow. The reason for this potential uncertainty is, a `CounterKind` is incremented
+//      via StatementKind::Counter at the end of the block, but (as in the case in this test),
+//      a CounterKind::Expression is always evaluated. In this case, the expression was based on
+//      a `Counter` incremented as part of the evaluation of the `if` expression, which was
+//      executed, and counted, 4 times, before reaching the overflow add.
+
+// If the program did not overflow, the coverage for `might_overflow()` would look like this:
+//
+//     4|       |fn might_overflow(to_add: u32) -> u32 {
+//     5|      4|    if to_add > 5 {
+//     6|      0|        println!("this will probably overflow");
+//     7|      4|    }
+//     8|      4|    let add_to = u32::MAX - 5;
+//     9|      4|    println!("does {} + {} overflow?", add_to, to_add);
+//    10|      4|    let result = to_add + add_to;
+//    11|      4|    println!("continuing after overflow check");
+//    12|      4|    result
+//    13|      4|}