This is the mail archive of the
mailing list for the Cygwin project.
Re: Release directories (was Re: [PATCH] setup: port to 64-bit, part 1)
On Thu, 14 Mar 2013 10:41:47 +0100, Corinna Vinschen wrote:
> On Mar 13 21:01, Yaakov wrote:
> > Before we do that, I think we need to consider a bit of
> > reorganization. As in any binary distribution, there are many "noarch"
> > packages which could be used for both i686 and x86_64. Providing two
> > identical copies is just a waste of storage and bandwidth for
> > sourceware, mirrors, and users.
> > Instead, I think it would make sense to make three sibling trees, one
> > for i686 (the current release/ directory), one for x86_64, and one for
> > noarch packages. Then, there would be two scans by upset: setup.ini
> > from i686 and noarch, and setup64.ini from x86_64 and noarch.
> Yes. You're right of course. This problem raises a few questions.
> - How do we store the packages on sourceware?
> Probably the easiest is to split into three dirs, as you suggested.
> The naming is pretty irrelevant, but it might be best to use a
> target name as the directory base, as on Linux:
> Alternatively we could stick to the current "release" name for the
> i686 distro and use only new, parallel dirs for noarch and new targets:
The former option may be easier to incorporate into cygport (e.g.
release/$ARCH/$NAME instead of dist/$NAME, and maybe a
user-configurable location for that release folder where all packages
would land). The latter option would be less of a burden on sourceware,
since only some of the packages would be moving instead of all of them.
> Another problem is to move the existing noarch packages into the
> right dir when we start. Well, at least this only has to be done
> once, baring any mistakes.
> - For uploading packages it's important to know where the new package
> has to go. Therefore, IMHO, it would make sense to change to a new
> package naming scheme, preferedly compatible with the versioning
> mechanism in upset, supported by cygport and easily recognizable by
> uploaders or upload scripts.
> Linux distros typically use the architecture after the version number:
> However, for backward compatiblity with the current mechanism, would it
> make sense to reorder it for Cygwin packages like so:
Nack. IIUC this form would confuse upset/setup/cygcheck to no end.
> - Do we have to change the RFU rules to include always the arch?
> If we change the naming convention of packages to include the arch,
> probably not.
Or if we make cygport organize things as above.