On Wed, Dec 10, 2014 at 3:57 PM, David Stone
I see no reason to allow code that is always a bug.
There are two possibilities here:
1) The user is expecting the object of that type to be there, but used the wrong type. This is the example Matt Calabrese gave: the variant contains long but I accidentally used int. This is a bug.
2) The user is wondering whether the type is there and is trying to get it if it is. Perhaps they are checking multiple types in sequence. However, this is a performance bug if the compiler cannot remove the "never going to work" code. Even if the compiler can remove the code, people reading it might not be able to do so mentally. Now you have code that will never execute, but all readers of your code assume it will (or why would you have written it)? I have worked on code bases with lots of code like this, and I have never once appreciated the misleading rabbit holes this sends me down during debugging.
Compilers don't disallow if (0) and I think this construct could occur in (generic) code. Could the same be true for the variant case? I.e., is it always a bug? -- Olaf