the .tar/.tgz/.tar.gz wars (was: Re: some try-ark suggestions)

Steve T comments:

> Would it be reasonable for a team's policy to *include* the "from
> where"?  Don't most shops generally work that way anyway?  For
> instance, I've often worked in shops where the (unwritten, but
> generally stated) policy was one of these:
> 
> 	"build everything from source obtained from the author's site
> 	only"
> 
> 	"only get stuff as binary packages from sunfreeware.com or
> 	redhat.com; we aren't in the business of maintaining a build
> 	environment" 
> 
> 	"get whatever you can find from any old mirror, doesn't
> 	matter, whatever...".  

Yes, I guess I've always kinda implicitly done the
first. (Or used a known trusted mirror.)  I kinda sloppily maintain
<pristine-homepage> and <pristine-source> attributes for
packages, e.g.

<pristine-homepage>http://www.gnu.org/</pristine-homepage>
<pristine-source>ftp://alpha.gnu.org/gnu/gzip/</pristine-source>

but these aren't particularly used yet.  (NB:
updates/additions always welcome.)

Berto, all this stuff is more brainless than you think.  The
sidai:ALL proto-package (sidai/package/ALL.xml),
`assemble-source' method, defaults the PKG_TGZ_FILE param to
<pkgname>-<ver>.tar.gz.  That's what most packages use.

*If* the original site for a package distributes it with
another name (or if it unpacks into something other than
<pkgname>-<ver>), then you will find <param>-setting code
that makes it the right thing for that package.

For example, Apache is different, so its proto-package says:

  <param name="PKG_TGZ_FILE">@package:ark-dirs:PKGS_SRC@/apache_@param:[email protected]</param>
  <param name="PKG_UNPACK_DIR">apache_@param:PKG_VERSION@</param>

In this way, the Right Thing usually happens.

If you, at your own site, stash your sources with some other
naming scheme, you can set PKG_TGZ_FILE in your own package
stuff.  (Not recommended, but you can.)

Gzip is a bit weird, because it is traditionally distributed
either as .tar or as .shar.

To make any of the above cleverer, just hack on the Bourne
shell script in an <assemble-source> method; even better,
hassle Harlan about the <get-source> method, which will go
grab the source off the 'net for you :-) [unwritten]

We probably *should* make the <assemble-source> method for
gzip more clever.  I am not gripped with a great idea at
this precise moment...

One reason I am reluctant to get too clever about this
ungzipping/untarring, though, is I would like the
tarball-that-was-used to be readily apparent, i.e. I want it
to be easy to have a webpage that says, "This package was
built from <that tarball over there>", with the info from
ARK-land.  If that exact tarball was not specifically
specified, but came out of the bowels of some shell-script
logic, then that's harder to do.

Again, there's kind of a philosophy thing here:  Your ARK
.xml files are supposed to express "this is what I used to
and how I built my site", not "this is some stuff I got from
sourceforge that managed to guess what I should use to build
my site".  Hmmm...

Will