Les Mikesell wrote:
On Thu, 2006-04-27 at 22:52, Mike McCarty wrote:
Here's one which may make more headway for understanding. There is
in any piece of software's life a thing called "integration test"
where it is tested with other pieces it must coexist with. Until
any given piece has been tested with other pieces, it may work
fine, or it may not, but we don't know.
That would imply that you shouldn't expect software to work
unless it all comes from a single vendor, which should
not be the case at all - and if it is, you probably
shouldn't use any of it. Your interfaces either do
what they are documented to do or not, and no amount
of changes to other parts should change that.
I guarantee you, with anything as complex as an entire
OS distribution, the interfaces do not all do exactly
what they are documented to do. Anyone believing otherwise
lives in a fantasy land.
Oh, I believe a lot of things are broken. I'm just not
sure I agree that replacing everything at once with
a new set that passed a couple of tests together is
the right way to deal with it. In particular it doesn't
make any sense to me to have to replace a kernel and
device drivers that have worked flawlessly on my hardware
for the last year or more with some wildly different
version just to get a new version of evolution and
some other desktop apps.
I don't recall making that assertion.
It is true that not all software would benefit as much
from integration test as others. OTOH, things like the
compilers, linkage editors, link libraries, kernel,
server programs, device drivers, installers, disc editors
and partition managers, and boot loaders all need some
integration test.
You seem to be forgetting that with fedora, the users
*are* the integration test. And for people who have their
own apps, the OS distribution is just supposed to supply
working unix-like interfaces.
I agree with you in some respects that Fedora Core is
more like a beta release than anything else. So I'll
concede that you have a certain point here.
But what I saw advocated, as I understood it, was the
complete abandonment of the concept of a release, and
instead piecemeal releases of all parts of the distro.
This would be unmaintainable. The installer, for example,
must be intimately aware of what the kernel and boot
loader expect. Trying to make it backwards and forwards
compatible would be... well I can't think of words
which would describe the mess.
Certainly, one could probably release Mozilla, or Thunderbird,
without much worry of incompatibilities. But not so
the kernel, or the compilers/devel tools, or the installer.
[snip]
Look at all the stuff SELinux broke.
Well, yes - the unix-like interface was pretty well
designed by about sysvr4 when linux started the attempt
to emulate it.
That's my point. It makes no sense for things to become
"unbundled" beyond a certain point. Some things *must*
be tested as a whole package, a release.
Mike
--
p="p=%c%s%c;main(){printf(p,34,p,34);}";main(){printf(p,34,p,34);}
This message made from 100% recycled bits.
You have found the bank of Larn.
I can explain it for you, but I can't understand it for you.
I speak only for myself, and I am unanimous in that!
