Re: Midori dependencies problem with sid
On Fri, May 29, 2009 at 02:23:31PM -0500, "Boyd Stephen Smith Jr." <bss@iguanasuicide.net> was heard to say:
> In <[🔎] 20090529013505.GA12317@emurlahn.burrows.local>, Daniel Burrows wrote:
> > Adding versioned Provides would affect all the software that tries
> >to process Debian packages and reason about their dependency
> >relationships. That's a lot of software, and all of it would be broken
> >until it learned about the new feature.
>
> Not completely broken. I imagine versioned Provides would be rather rare.
Yes, but as soon as one was added to the Packages file, everything
that parses it would break. Some software might fail "gracefully",
ignoring the versioned provides, but I would not count on this at all.
> Also, shouldn't most of that software be using libapt? If someone wanted to
> implement it, how could they get a list of software to fix?
I don't know what tools like britney and dak use. As far as I know,
the software that runs on the buildds hand-parses the Packages files and
would break. I also would not be surprised if there are lots of tools
users have written to hand-parse Packages files.
I'm also a little leery of having multiple unrelated providers of a
versioned name -- programs that require a versioned name are usually
really requiring some behavior that was tied to a particular version,
and unless you have programs that behave in truly identical ways, it
seems likely that you would be bitten by versioned provides. The cases
where it would be safe to use them seem narrow enough to me that we can
just use dummy packages, normal virtual packages (e.g., create a
"bash-3.0" package and nail down what that means), etc.
Daniel
Reply to: