| Issue |
161333
|
| Summary |
Lifetime safety can timeout on cyclic substitution in a loop
|
| Labels |
clang:temporal-safety
|
| Assignees |
|
| Reporter |
usx95
|
```cpp
struct MyObj { int id; ~MyObj() {} };
void long_cycle_4(bool condition) {
MyObj v1{1};
MyObj v2{1};
MyObj v3{1};
MyObj v4{1};
MyObj* p1 = &v1;
MyObj* p2 = &v2;
MyObj* p3 = &v3;
MyObj* p4 = &v4;
while (condition) {
MyObj* temp = p1;
p1 = p2;
p2 = p3;
p3 = p4;
p4 = temp;
}
}
```
Current benchmark results:
Test Case: Pointer Cycle in Loop
**Timing Results:**
| N (Input Size) | Total Time | Analysis Time (%) | Fact Generator (%) | Loan Propagation (%) | Live Origins (%) |
|:---------------|-----------:|------------------:|-------------------:|---------------------:|------------------:|
| 50 | 56.10 ms | 79.14% | 0.00% | 78.77% | 0.00% |
| 75 | 156.51 ms | 91.45% | 0.00% | 91.24% | 0.00% |
| 100 | 344.89 ms | 94.44% | 0.00% | 94.32% | 0.00% |
| 200 | 2.47 s | 98.46% | 0.00% | 98.42% | 0.03% |
| 300 | 8.85 s | 99.16% | 0.00% | 99.14% | 0.01% |
This is also found to be a scaling issue for Polonius: https://github.com/rust-lang/rust/issues/140454
Since we do not aim for rigorous memory safety for C++ using Lifetime Safety, we can afford to cut off the fixed point iteration after some iterations. (See https://github.com/llvm/llvm-project/issues/154509)
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
