Subject: Re: Names of the patch files
To: NetBSD Packages Technical Discussion List <tech-pkg@NetBSD.org>
From: Johnny Lam <firstname.lastname@example.org>
Date: 04/19/2006 17:25:03
> On Wed, Apr 19, 2006 at 10:20:22PM +0200, Georg Schwarz wrote:
>> * When a new patch is created, it is to be named as to fill existing
>> gaps in the naming order created by previously removed patches as
>> long as such gaps exist (at least that's how I think it has been done
>> so far)
> This is not always a good idea. If you want to have a patch included
> both in HEAD and stable, it makes a lot sense to use the same name.
When I add patches, I actually check the Attic to see if any previously
used patch-[a-z][a-z] had patched that file in the past, and if so, then
I resurrect that patch name.
As a pkgsrc developer, my main concern with working with patches is
being able to track how each source file is being patched across package
updates without needing to inspect every patch's CVS history. For this
reason, I actually think we should go with one big patch file
per-package that is the concatentation of all of the single-file patches
we currently have. The current naming scheme is just an artifact of the
original FreeBSD ports import, and even FreeBSD ports has dropped the
patch-[a-z][a-z] convention for quite a while now. If there was an
automatic way to name patches based on the filename being patched, that
would also solve the problem I mentioned, but I notice that the patch
names tend to get incredibly long and ugly.
To flesh out the "jumbo patch" idea, we would create a make target that
explodes the jumbo patch for the benefit of developers so they can
selectively apply individual sub-patches. We would also create a make
target that concatenates the patches into one big jumbo patch in a
consistent way (probably based on sort order of the file being patched),
which can then be committed to pkgsrc.
-- Johnny Lam <email@example.com>