On 18/05/10 20:19, retard wrote:
What if I'm using a clean room implementation of D with a custom backend
and no accompanying C compiler, am I not allowed to compare the
performance with anything?

When people compare C compilers, they usually use the latest Visual
Studio, gcc, icc, and llvm versions -- i.e. C compilers from various
vendors. Using the same logic one is not allowed to compare dmc against
those since it would always lose.

I don't believe Walter is arguing against this methodology. What he is arguing against is comparing dmd with gcc for example. Comparing ldc with gdc and dmd is fine, comparing dmd with dmc is fine, but when it comes to comparing D and C, he believes you should compare compilers using the same backend, that is dmd and dmc rather than dmd and gcc. Or that's what I took from it.

This said, I don't agree with that methodology, unless it's only a small test. If you're comparing lots of C compilers and D you should include dmc for example if you're using dmd as the D reference, or clang if you're using ldc as a reference. If you're comparing C and D, you should stick to compilers with the same backend, otherwise the one with the superior backend will always win, and it's not a fair interlanguage comparison.

Reply via email to