Compiling Fedora... athlon

[Owen Taylor]

On Tue, 2004-02-10 at 01:42, Ivan Gyurdiev wrote:

> I've submitted bugzillas for several of those (freetype-headers for 
> gnome-print, the tcsh bug, desktop-printing, some of the gnome 
> deprecated bugs ... but I prefer mail as the means for communication. 
> Plus since a bug stayed in rawhide bugzilla for a year and a half
before 
> I got a reply, I just don't trust the thing anymore.

I think that's an excellent reason to trust bugzilla! If you sent
mail to a mailing list, and someone didn't have a chance to look
at the problem for 18 months, they'd have completely forgotten
that your report existed...

I figure I should follow up here since I added a comment to 115081:

 I can only answer for my own packages (fontconfig, gnome-print,
 libgnomeprint of the above), but for those, I'm not sure filing
 such bugs is valuable. I still wouldn't make the change until I
 had to rebuild the package for some other reason, and at that
 point, I'm going to be reminded by the build system to fix this...

Based on my current (mostly upstream GTK+) workload, I don't have
a lot of flexibility in that at the moment, but I think it is an
interesting policy question for Fedora in general - just how important 
is keeping the set of RPMS always recompilable with the current
tree?

Pro:
 - People seem to be rebuilding RPMS a lot; if something breaks,
   bug reports appear very quickly.
 - Makes mass rebuilds to check, say, compiler changes, easier  
   if everything builds all the time.
 - Generally good hygiene

Con:
 - It takes developer time that could profitably be used elsewhere
 - Making everything fully bootstrappable involves significant
   engineering ... e.g., fixing the freetype <=> XFree86 build
   loop dependency would require rewriting the makefiles for
   freetype-demos and getting the changes upstream. (*)

Right now, I tend to consider it a "would be nice" but not a
"must"; but if it turns out to be an important goal for the Fedora
project than it probably needs to be enforced in the process,
say by not shipping betas until all packages pass a mass-rebuild.

Regards,
						Owen

(*) Note that a bootstrap and mass-rebuild are different things:

   bootstrap: Recompiling everything from the ground up;
    clearly requires using at least a few packages from a previous
    build or a cross-compiler.
   mass rebuild: Rebuilding all packages, but just against the
    current tree.