On Thu, 21 Oct 2004 12:48:53 +0100, David Dorward <[EMAIL PROTECTED]> wrote:
> USE flags specify (mostly) what optional packages > to compile against. For Debian to do this, all packages would have to > be compiled against all combinations of all optional dependancies. As > it is, without that level of choice, Debian is already too big to fit > on a DVD! (And that doens't begin to cover the additional level of > testing that would be required). I've been meaning to get some stats for some time - but I'd be interested to see how feasible it would be to package up the binary .o files for a package and the linking stage would be done locally - advantages: faster than compiling all from scratch for user. Assuming that a program built with configuration X and configuration Y share a significant subset of .o files; the package could include the superset (so, you download the .o files, and link according to your preferences e.g. X support). Dynamic dependencies so if you don't want emacs with x support, you don't pull in the x libraries, but the mirror still only has one X package (albeit with spare .o's for the two configurations). Disadvantages: the linking stage means slower than a straight binary download. Slightly larger packages (well, that all depends on how many .o files the configuration change would impact) An alternative would be a subset of common .o's and a small compile job for the user. I guess gentoo could reach this stage just by having a distribution-wide distributed ccache, or something. -- Jon Dowland [EMAIL PROTECTED] -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
