Dawid Ciężarkiewicz schrieb:

> BTW. Why to keep varadic arguments in D? I'm just courious. Wouldn't be good
> to let it by used with "C" linkage only. IMHO banning it let compiler to
> use PASCAL like function returning with asm { "ret <number>" } which I read
> is more efficient. I'm not the expert, but I always thought that variadic
> arguments lays in C++ becouse of C compatiblity. Maybe D should leave it.

In fact, it's not really used in D - and those funcs in question are C library functions.

But why drop it? It's as crippled as in C. And we don't have a good replacement for it yet, such as powerful array literals. And there probably won't be any problem adding another varargs system later, as the time comes - they can coexist well.

In C, varargs affects performance, because the possible ambiguity requieres that all functions have the same calling convention, which happens to be a caller clean-up. C++ and D requiere that the functions be declared before use, thus there is no ambiguity and the compilers choose a calling convention depending on the function prototype - for varargs functions caller clean-up is used, otherwise the more efficient callee clean-up with possible register use.

On the other hand, would we loose anything by disallowing varargs in "D" convention right now?

-eye