[Bug tree-optimization/126848] New: Repeated reads of an indeterminate value do not yield the same value

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126848

            Bug ID: 126848
           Summary: Repeated reads of an indeterminate value do not yield
                    the same value
           Product: gcc
           Version: 16.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: peppe at gcc dot gnu.org
  Target Milestone: ---

Consider this testcase:

```
[[gnu::noipa]] 
int g(int c) 
{
    int a;
    int x = c ? a : 1;
    return x - a;
}

int main() 
{ 
    return g(1);
} 
```

On GCC 16.2 with -std=c++26 -O2, this program exits with 1. 

Shouldn't it exit with 0? It is my understanding that, under P2795R5 / C++26
rules (implemented in GCC 16), re-reading the same indeterminate value is
indeed erroneous behavior, but it must still yield the same value back:
https://eel.is/c++draft/basic.indet#example-1

This means that `x-a` must be 0 if `c` is true, no matter what's the erroneous
value of `a`.