On Mon, May 18, 2009 at 3:29 PM, Robert Fraser <fraseroftheni...@gmail.com> wrote: > Jarrett Billingsley wrote: >> >> it'd be an alternative to CTFE (shudder) > > Why shudder? CTFE has familiar syntax (the syntax of the runtime language) > and, I've found, less bugs in general.
It's extremely difficult to make CTFE functions that work. The compiler often coughs on seemingly-legal code, sometimes due to bugs. The biggest issue with developing CTFE functions is that if something does go wrong during CTFE, all you get is a message along the lines of "function X can't be evaluated at compile time", with no indication as to _why_ it couldn't, no stack trace etc. There's also no means of outputting debugging statements during CTFE. At least with templates I have pragma(msg). CTFE is also mainly useful for building up string mixins, but since everything inside a CTFE function has to be code that can be executed at runtime as well, you forgo some really useful metaprogramming features that are native to templates, like pattern matching. So it ends up being convenient in simple cases (i.e. given a list of names, generate a bunch of almost-identical declarations for them), but not in more complex ones. Not to mention it's much slower than template instantiation and eats memory for lunch.
