Why does C++ need throw()

Quote:

---

Originally Posted by brewbuck

That's exception safety, not exception specification.

---

That was sort of my point. The design of the STL had nothing to do with avoiding or encouraging exception specifications.

Quote:

---

Originally Posted by brewbuck

You don't need exception specifications -- ever.

---

Such sweeping statements clearly demonstrate vast experience and knowledge.

The issue, as cpjust has said, is that, in C++, developers often assume that exception specifications behave in a different manner from what they actually do. Which means that developers need to better understand what exception specifications in C++ mean, or that the standard needs to be modified to comply with programmer expectations.

The latter is an impossibility, of course, because you can't introduce such an incompatibility. Also, the same reasons that made exception specifications what they are and not what programmers (especially Java programmers) expect still apply. Finally, exception specifications such as Java's may not be the best idea either; at least, many Java programmers (such as the Spring framework design team) and the designers of C# think so.

As for educating programmers, that would be possible, but to what end? Here's the effect of a throw statement: it transforms

Code:

---

void foo() throw(A, B)
{
  // Code
}

---

into

Code:

---

void foo()
try {
  // Code
} catch(A &) {
  throw;
} catch(B &) {
  throw;
} catch(...) {
  try {
    std::unexpected();
  } catch(A &) {
    throw;
  } catch(B &) {
    throw;
  } catch(...) {
    std::terminate();
  }
}

---

unless the exception specification explicitly includes std::bad_exception, in which case the std::terminate() call is replaced by throwing a std::bad_exception.

The obvious problem is: how is that useful?

Quote:

---

Originally Posted by grumpy

Such sweeping statements clearly demonstrate vast experience and knowledge.

---

I don't see how sarcasm helps. There is no problem that could be solved in C++ more elegantly or optimally by introducing an exception specification. It was from the beginning a weak attempt to enforce what kinds of exceptions can leave a function. If it could actually do that, it would be useful. But unlike in Java where you get a BUILD ERROR if you break the exception specification, C++ just explodes if you violate it.

That is not useful, helpful, or anything else. It deludes the programmer to thinking that the specification is somehow making his code more robust -- "I've specified no throw, so surely if I call this function, no exception will be thrown." It's simply wrong, and it makes the entire mechanism UTTERLY POINTLESS.

Quote:

---

The issue, as cpjust has said, is that, in C++, developers often assume that exception specifications behave in a different manner from what they actually do. Which means that developers need to better understand what exception specifications in C++ mean, or that the standard needs to be modified to comply with programmer expectations.

---

The programmer expectation is usually that the specification will be enforced at compile time. That's simply impossible to accomplish in C++.

Quote:

---

Originally Posted by grumpy

Not true. The STL is designed with defined levels of exception safety in the presence of any exception: for example, if an operation on an element of a container throws an exception, then the container remains in a sensible state (it can be destroyed safely, does not generate memory leaks, etc) and the exception propagates to the caller. There are a few conditions on this (eg destructors of container elements are not allowed to throw exceptions, elements of containers must also keep themselves in a destructible state if an exception is thrown, etc). The net effect is that the STL behaves in a rational manner if any exception is thrown, and does not need to know anything about exceptions which might be thrown.

---

That's all true. I was just saying that the STL does't use throw declarations in function prototypes. cpjust sais it does, and he's probably right.

Quote:

---

Originally Posted by King Mir

That's all true. I was just saying that the STL does't use throw declarations in function prototypes. cpjust sais it does, and he's probably right.

---

Or more likely, it's implementation dependent.