Thread: Assertion failure taking place of a try-catch

I know I can't create this iterator. What is not clear to me is why the try-catch block is failing to isolate the error and instead I get an assertion failure at runtime.

(note: vec1 has been properly declared and initiliazed)

Code:

/* ... */
#include <stdexcept>
/* ... */
try {
    vector<int>::const_iterator ivec = vec1.begin() - 1;
} catch (runtime_error err) {
    cout << err.what();
}

Initialize your iterator with begin(). The reason your try/catch block isn't working is because there was not a runtime exception: you tried to initialize an iterator with some random place in memory which is an out_of_range exception, I think.

Well, the range_error class is derived from runtime_error. So it should be caught with the main class, no? I do have the same problem if I try to catch it with range_error, mind you.

I don't want to initialize this iterator correctly. I'm curious now as to why the assertion. My first guess is that this is a variable definition and I couldn't catch it. But moving the definition outside of the try block didn't solve the assertion.

> Well, the range_error class is derived from runtime_error.
No, out_of_range isn't a runtime_error.

Well, the range_error class is derived from runtime_error.

citizen referred to out_of_range, which is derived from logic_error.

Ah. I understand now.

Well... I did some further testing with out_of_range and the logic_error base class. To no avail. I'm not sure if this is an implementation issue. MinGW doesn't even report an error. I'm able to output the dereferenced iterator and even increment or decrement it inside the try block without any error being caught. Whereas VC++ 2005 Express throws the assertion.

Now that I think about it, why would an exception be thrown from begin()? It's not implemented like that. If you want to catch stuff in your for loop you'd have to throw it yourself. Something like:

Code:

try {
          vector<int>::const_iterator it = vec1.begin() - 1;
          if( *it != vec1.begin() )
             throw logic_error();
          // for goes here
// I've not tested this but it should be okay...

Be advised I've always had trouble throwing out_of_range exceptions manually, so maybe just throwing a logic error is better. Kurt is still right by the way. It's where its at()

Cheesy pun.

Yes. That's it. Thanks all