1| |#![allow(unused_assignments)] | |
2| |// expect-exit-status-101 | |
3| | | |
4| 4|fn might_panic(should_panic: bool) { | |
5| 4| if should_panic { | |
6| 1| println!("panicking..."); | |
7| 1| panic!("panics"); | |
8| 3| } else { | |
9| 3| println!("Don't Panic"); | |
10| 3| } | |
11| 3|} | |
12| | | |
13| 1|fn main() -> Result<(), u8> { | |
14| 1| let mut countdown = 10; | |
15| 11| while countdown > 0 { | |
16| 11| if countdown == 1 { | |
17| 1| might_panic(true); | |
18| 10| } else if countdown < 5 { | |
19| 3| might_panic(false); | |
20| 6| } | |
21| 10| countdown -= 1; | |
22| | } | |
23| 0| Ok(()) | |
24| 0|} | |
25| | | |
26| |// Notes: | |
27| |// 1. Compare this program and its coverage results to those of the similar tests `abort.rs` and | |
28| |// `try_error_result.rs`. | |
29| |// 2. Since the `panic_unwind.rs` test is allowed to unwind, it is also allowed to execute the | |
30| |// normal program exit cleanup, including writing out the current values of the coverage | |
31| |// counters. | |
32| |// 3. The coverage results show (interestingly) that the `panic!()` call did execute, but it does | |
33| |// not show coverage of the `if countdown == 1` branch in `main()` that calls | |
34| |// `might_panic(true)` (causing the call to `panic!()`). | |
35| |// 4. The reason `main()`s `if countdown == 1` branch, calling `might_panic(true)`, appears | |
36| |// "uncovered" is, InstrumentCoverage (intentionally) treats `TerminatorKind::Call` terminators | |
37| |// as non-branching, because when a program executes normally, they always are. Errors handled | |
38| |// via the try `?` operator produce error handling branches that *are* treated as branches in | |
39| |// coverage results. By treating calls without try `?` operators as non-branching (assumed to | |
40| |// return normally and continue) the coverage graph can be simplified, producing smaller, | |
41| |// faster binaries, and cleaner coverage results. | |
42| |// 5. The reason the coverage results actually show `panic!()` was called is most likely because | |
43| |// `panic!()` is a macro, not a simple function call, and there are other `Statement`s and/or | |
44| |// `Terminator`s that execute with a coverage counter before the panic and unwind occur. | |
45| |// 6. Since the common practice is not to use `panic!()` for error handling, the coverage | |
46| |// implementation avoids incurring an additional cost (in program size and execution time) to | |
47| |// improve coverage results for an event that is generally not "supposed" to happen. | |
48| |// 7. FIXME(#78544): This issue describes a feature request for a proposed option to enable | |
49| |// more accurate coverage results for tests that intentionally panic. | |