I've been reading the discussion ...
Firstly, I'm glad to have disposed of the `byte-for-byte original
source archive' idea (and that some of the people I thought were
advocating this were merely advocating that we should be able to
extract the original source files somehow from our source archive,
perhaps in a differently named directory or some such).
It seems to me that the `we want a single file' idea is perhaps
starting to be a problem, and that some of the alternatives people
have suggested may have some merit.
I'll therefore describe to you an alternative proposal to the one I
described last weekend (the one with a single `ar' archive). I'd like
to see more discussion about this, as I'm still not convinced that all
the possible issues have been raised.
Please don't bother having long flamewars about the merits of one
thing versus another if you're just trying to convince the other
people of your correctness. It's me and Bruce you need to convince, I
think :-).
Here's the alternative I've dreamed up:
* The source package is three files:
- hello_1.4-3.dsc
DSC= Debian Source Control (suggestions for better extensions
welcome). This contains a dchanges/debian.control-like format, for
example:
Source: hello
Version: 1.4-3
Architecture: any
Depends: gcc { standard dpkg control file syntax }
Generates: hello { names of binary packages, comma-separated }
Data-MD5sums:
faa56f7d564b1972f66a2d17ddf97413 hello_1.3-4.diff.gz
d2cb670eee141fc08eaa4a794b8b68fe hello_1.3.orig.tar.gz
This would optionally be PGP-signed. The `Architecture: any' means
that the source is arch-independent, but the binary isn't.
`Architecture: all' would mean it was a truly arch-independent
package (documentation, for example).
- hello_1.3-4.diff.gz
The Debianisation diff, as we have now.
- hello_1.3.orig.tar.gz
The original source code, reorganised if necessary into a tarfile
that unpacks into a subdirectory called hello-1.3 or perhaps
hello_1.3.
* Issues:
- Perhaps we need to think again about these hyphens in the context
of source packages, so that we can distribute GNU source tarfiles
unchanged.
- Uploading a new package which only has changes to the diff becomes
trivial. You have to supply a new .dsc and a new .diff.gz, but you
can just name the old original source in the .dsc.
- We need a dpkg-source script to unpack the tarfile and apply the
diff safely (and it can automatically change debian.rules to be
exectutable), but people can do it themselves if they want. Also,
we need a dpkg-source script which runs diff and checks that there
are no differences that diff can't handle (deleted files, retargeted
links, &c).
- We need to store the .dsc somewhere when the source is `opened up',
ie, made into a filesystem tree. Either this should be put
somewhere when dpkg-source unpacks an archive, or perhaps it should
have some canonical name in the Debianized source archive
(debian.sourcecontrol perhaps). The dpkg-source script can produce
the Data-MD5sums and Version fields, and check the Source field, so
only Soource, Architecture, Depends and Generates would be needed
here.
* As a reminder, the `one file' format I favour at the moment is an `ar'
archive with members for the control file, diff, original source
tarfile and optionally signature.
Ian.
