Subject: Re: Kernel compile 100 times faster (this sound provocating,doesnt it?:)
To: None <chammer@phyd2hammer.uni-bielefeld.de>
From: Mark W. Eichin <eichin@kitten.gen.ma.us>
List: current-users
Date: 02/03/1996 17:53:32
Interesting idea - but I'd like to see what numbers you've got that
indicate that it would help at all. Preprocessor time is negligible
compared to cc1 time, especially in the optimizer... same goes for
memory use.

What might *really* help is some further work on the "precompiled
headers" idea which already has some hooks in gcc. The hard part is
that you need to actually save a bunch of the cc1 state to get any
savings, and come up with a database mechanism to find it.

To get any benefit, though, you need to have common set of initial
includes in your source files; once you've done this, you reap
*another* benefit, at least on a.out, of link-time compression of
debugging symbols (I don't know if NetBSD is using a linker that
supports that though. The idea is simple, and is fairly effective with
a.out, and is in the latest GNU binutils; it doesn't require any
format change, just recognizing that if two STABS strings are the same
you can use the same offset into the file instead of putting in two
copies of the string... the include ordering has the effect of causing
sequentially allocated type numbers to be the same, yielding more
reusable strings.)

			_Mark_ <eichin@kitten.gen.ma.us>
			The Herd of Kittens