random comments about secondary arch proposal

[David Woodhouse]

On Sat, 2007-06-16 at 14:08 +0200, Ralf Ertzinger wrote:
> On Fri, 15 Jun 2007 12:38:28 +0100, David Woodhouse wrote
> > You're right -- as things stand, we do need a team of folks who can
> > help out with basic issues where the packager can't cope. It usually
> > isn't arch-specific; very little really is when you get down to it.
> 
> Arch specific in the sense of "this does not work on PPC": yes, I can 
> agree with that.

That's what I meant, yes.

> Arch specific in the sense of "does work on i386 and nowhere else":
> it seems to me that there is quite an amount of C code that assumes
> that an int is 32 bit and the world is little endian.

Code written _that_ badly is rarely good enough to be shipped with the
'Fedora' label on it anyway, for a variety of reasons.

It's also a case which I was mostly ignoring as uninteresting in this
context, since it's the "needs porting" case which just gets a one-off
ExcludeArch; either when the package is first imported or when the arch
team build all packages for their arch before applying to become a
Secondary Architecture. It hardly takes any effort for the package owner
at all; certainly not any recurring effort.

The _important_ case we have to care about is when a package _used_ to
build and suddenly doesn't. I think that's the one that people are
concerned about. Perhaps because they haven't actually looked at the
data and didn't realise that a significant proportion (and probably even
the _majority_) of such failures aren't actually arch-specific bugs at
all, when you get down to the root of it.

And even the failures which _are_ arch-specific can easily be handled
with a retrospective ExcludeArch and the 'ship it anyway' button.

-- 
dwmw2