Talk:cpp/error/uncaught exception
The description states that "Sometimes it's safe to throw an exception even while std::uncaught_exception() == true", but the example - "if exceptions are caught and ignored in a destructor, they can't propagate out of it and won't lead to std::terminate" - doesn't seem to relate to the assertion. Is it unclear or unrelated? 92.235.157.64 13:40, 22 November 2012 (PST)
- It's exactly the case when it is safe: suppose stack unwinding is in progress and a destructor of some local object is called. In the body of that destructor, there is a call to a potentially throwing function, surrounded in a try/catch(...). The destructor calls that potentially-throwing function, and inside that function, uncaught_exception() == true, but it's safe to throw. I suppose it's unclear, tried to edit for clarity. --Cubbi 19:16, 22 November 2012 (PST)
- Ahh yes, didn't think of being inside a try..catch in a destructor during stack unwinding. Silly brain. Thanks for the edit - clarification works fine. 92.235.157.64 15:47, 24 November 2012 (PST)
[edit] Accuracy detection of stack unwinding durning destruction
Hello everyone!
I suppose that class Foo from the sample in the article cannot correctly detect if there is stack unwinding during destruction.
To prove myself I make a little demo, that shows two cases where this class was correct and two more cases where it wasn't.
https://godbolt.org/z/GEaYn6Mcz
And I think it's not so synthetic example and it can be met in a real program.
And because we can make a very synthetic function that can run any code with any value of uncaught_exceptions() (see example 5 in code from the link above), I suppose that we cannot use a result of uncaught_exceptions() to make any predictions about stack unwinding in the general case.
I suppose that Foo detection works only if it was created on the stack without moving or copying.
- the example in the page is demoing the use case this function was designed for, as found in boost.log (ok, they use >= instead of !=, guess we could too). And I see boost.leaf got it, too. Yes indeed it can be subverted. Nice demo. --Cubbi (talk) 20:08, 6 August 2021 (PDT)
