binary patches?

[Danny Pansters]

On Thursday 15 March 2007 00:00, Gary Kline wrote:
> On Wed, Mar 14, 2007 at 05:07:43PM +0100, Gabor Kovesdan wrote:
> > Gary Kline schrieb:
> > >	Regarding most (or many) of the port changes--say, upgrading
> > >	foo-2.1.9_5 to foo-2.1.9_6, if  the upgrade could be done by
> > >	downloading a binary diff file, could the resulting
> > >	/usr/local/bin/foo-2.1.9_6 be achieved by downloading a
> > >	relatively small binary patch?  Seems to me that smaller scale
> > >	upgrades could be done this way in preference to re-compiling
> > >	ports or downloading entire pacakes.  --Same would go for any
> > >	dependencies.
> > >
> > >	Why is this a bad idea!

I don't think it's a bad idea at all, but impractical

> > >
> > >	gary
> >
> > The final form of actual binaries depend on a lot of things, e.g. which
> > version of dependency you compiled with, which CFLAGS you have used,
> > what options the port you built it. Some of these applies to packages as
> > well, that's why I prefer ports over packages at all. E.g. let's see
> > lang/php5. It does not have the apache module enabled by default. If it
> > were, then the problem comes up with Apache versions. IIRC, 2.2 is the
> > default now, but what if you use 2.0? How would you install php for your
> > apache version from package? The situtation has been already pretty
> > complicated with packages if you have higher needs for fine tuning, but
> > you can use them if you don't have special needs. Binary diffs would be
> > so complicated that I think this way we could really not follow.

Yes, seperate binpatches for every port option or build setting. And for any 
differently configured dependency! And those would have to be all checksummed 
also. Preferably from a seperate reliable source. So it quickly gets much 
much bigger and complicated than a source-only approach. Which is complicated 
enough as it stands :)

> > If you need simplicity at all, use portupgrade with packages. It has an
> > option (don't remember which one) you can use to make it fetch packages
> > instead of building from source. Nowadays, this network traffic should
> > not be a real problem, I think.
>
> 	You've brought up a lot of things I didn't consider; this was
> 	part of the reason for my post.  It seems to me that there would
> 	need to be some simple ground rules from the binary patches I'm
> 	got in mind.  The *default* CFLAGS in the port would match those
> 	in the patch is one place to start.

But you only had an example with one single binary. Not many useful apps 
installs one single binary. And then there's a multitude of libs (which of 
course depend on other libs)

> 	Obviously, this could get way out of hand very quickly.  Two of

Yes, after two iterations or so. IOW: instantly.

> 	my slowest servers (one 400MHz, 192M RAM) were rebuilding parts
> 	of the KDE suite; the new kdelib-3.5.6 [??] just finished and I
> 	already scp'd it over to my more beefy platform.  Once I've got
> 	all my servers up to date, it may not be that hard to keep them
> 	current.  You're right that bandwidth isn't a problem--um, in
> 	most places {{ clearing my throat! }}.  Bandwidth isn't the main
> 	issue.  It's time.
>
> 	cheers!

Also, it frequently happens that when upgrading a large project (say Gnome) 
you pretty much have to remove the old-install the new anyway.

As others said as well, it's a nice idea if you have unlimited manpower, but 
practically speaking it's a maintenance nightmare much more than a source 
approach is.


Cheers,

Dan

> 	gary
>
> > Regards,
> > Gabor
> > _______________________________________________
> > freebsd-questions at freebsd.org mailing list
> > http://lists.freebsd.org/mailman/listinfo/freebsd-questions
> > To unsubscribe, send any mail to
> > "freebsd-questions-unsubscribe at freebsd.org"