svn commit: r227536 - in head: release share/man/man7

[John Baldwin]

On Thursday, November 17, 2011 12:43:12 pm Ken Smith wrote:
> On Thu, 2011-11-17 at 11:41 -0500, John Baldwin wrote:
> > On Thursday, November 17, 2011 10:11:36 am Ken Smith wrote:
> > > On Thu, 2011-11-17 at 14:57 +0000, Alexey Dokuchaev wrote:
> > > > On Thu, Nov 17, 2011 at 09:44:52AM -0500, Ken Smith wrote:
> > > > > This is the problem we are trying to "solve":
> > > > > 
> > > > > Supported TARGET/TARGET_ARCH pairs for world and kernel targets
> > > > >     amd64/amd64
> > > > >     arm/arm
> > > > >     arm/armeb
> > > > >     i386/i386
> > > > >     ia64/ia64
> > > > >     mips/mipsel
> > > > >     mips/mipseb
> > > > >     mips/mips64el
> > > > >     mips/mips64eb
> > > > >     mips/mipsn32eb
> > > > >     pc98/i386
> > > > >     powerpc/powerpc
> > > > >     powerpc/powerpc64
> > > > >     sparc64/sparc64
> > > > 
> > > > As I see it, for every pair except pc98/i386, second part should be 
used.
> > > > For pc98/i386, first (pc98).  Problem solved.  ;-)
> > > > 
> > > > ./danfe
> > > > 
> > > 
> > > I'd still sort of prefer no special cases.  However ...
> > > 
> > > For the ISO / memstick filenames we could just program in `uname -p`
> > > and ask the pc98 builder to modify the filenames post-build.  But
> > > we still have the dual names needed for the FTP site layout.  There
> > > it needs to be fully automated in the installer.
> > > 
> > > So, given it seemed like we're sort of stuck with having the dual
> > > names appearing in other places combined with it never causing us
> > > to have special cases and/or conflicts it seemed like just biting
> > > the bullet and having them in the ISO / memstick filenames too ...
> > > 
> > > Have I mentioned I don't like any of the options?  :-/
> > 
> > I think collapsing down to one name if uname -m == uname -p is not that 
> > terrible and would preserve the existing layout for most of the current
> > cases (only pc98 would change, yes)?
> > 
> 
> If you're referring to the FTP directory tree layout we wind up with:
> 
> 	.../releases/amd64/9.0-RELEASE
> 	.../releases/amd64/ISO-IMAGES/9.0
> 
> for an example of uname -m == uname -p.  But for our two powerpc related
> architectures we get:
> 
> 	.../releases/powerpc/9.0-RELEASE
> 	.../releases/powerpc/ISO-IMAGES/9.0
> 	.../releases/powerpc/powerpc64/9.0-RELEASE
> 	.../releases/powerpc/powerpc64/ISO-IMAGES/9.0
> 
> I'm not sure I like the inconsistency.

Given the available tradeoffs I prefer this to amd64/amd64.  We could also 
define the rule another way, which is if a given TARGET only has a single 
TARGET_ARCH you just use TARGET, otherwise you use TARGET/TARGET_ARCH.  (This 
can be parsed out of the output of 'make targets' fairly easily.)  That would 
let you have:

releases/amd64/9.0-RELEASE
releases/powerpc/powerpc/9.0-RELEASE
releases/powerpc/powerpc64/9.0-RELEASE

-- 
John Baldwin