I don't think this has to be perfect to be a big win over what we've got.
The worst slowdown anybody is reporting with Zack's simple-minded approach is on the order of a minute. That's on slow machines, where a GCC bootstrap will take hours. So, I'm not worried about the time *at all*. I'd be very happy to pay the cost to avoid having to keep dependencies up to date.
I'm not worried about the case where the header files are renamed screwing things up. That's rare. And there's a workaround: touch the header, and then rebuild the dependencies. If we can fix this, great, but if we can't, I don't see it as a showstopper.
We do need a story for generated headers. I'd be happy with explicit dependencies in the Makefiles indicating what source files depend on what generated headers. We'd still be able to get rid of 99% of the dependencies in our Makefiles, and avoid most of the cases where something either (a) fails to get rebuilt, or (b) gets rebuilt too often.
My two cents,
-- Mark Mitchell CodeSourcery, LLC [EMAIL PROTECTED] (916) 791-8304
