[phobos] enforce() vs. assert() for range primitives

[Jonathan M Davis]

The ideal situation would be to use asserts in all cases where it's going to be 
bugs in the program rather than bad user input and that they go away in release 
mode. The problem is the fact that Phobos is compiled as an external library 
rather than being compiled in with your code when you compile. And since there's 
no way (at least no standard way) to have a release and debug version of the 
Phobos libraries, with the release version being used in release mode and the 
debug version being used in debug mode, that pretty much means that you _always_ 
compile your code with the release version of Phobos. This makes assert 
_useless_ for users of the library. If you want checks beyond within Phobos 
code, you _have_ to use enforce(), and since enforce has issues with inlining 
and it doesn't go away in release mode, it's like forcing a portion of Phobos 
into debug mode.

_Ideally_, we would find some sort of standard way to make Phobos used in debug 
mode when compiling in debug mode and release mode when compiling in release 
mode. That way, asserts could be used in a lot of these cases since a lot of 
them really shouldn't be checked in release mode. Personally, I _really_ don't 
like the fact that the libraries aren't necessarily in the same mode as your 
code, though I'm not sure that there's a good way to deal with that other than 
having some standard naming scheme for debug vs release libraries which Phobos 
uses and dmd understands so that you get a debug version when compiling your 
code in debug mode, and you get a release version when compiling in release 
mode. Barring that however...

You'd almost have to take it on a case by case basis. In some cases, you're 
going to need checks and it would be bad not to have them, while in others, it 
would just cost too much to leave them in. It's a tough call. If anything 
though, I think that it's a prime example of needing to find a better way to deal 
with debug vs release and libraries. If we had a standard way to have both 
available, then dmd could use both with user code, and then we can use asserts 
like they should be used. As it stands though, we're probably going to have to 
use enforce() where we _need_ checks for safety and asserts otherwise. But since 
using asserts is almost pointless.... Bleh. It's just ugly.

- Jonathan M Davis