Re: Question about 'provides' and emacs packaging

[Jon Turney <jon.turney@dronecode.org.uk>]

On 06/10/2021 13:01, Ken Brown via Cygwin-apps wrote:
> On 10/5/2021 2:24 PM, Achim Gratz wrote:
>> Ken Brown via Cygwin-apps writes:
>>> There are currently five emacs packages: emacs-common, emacs,
>>> emacs-X11, emacs-w32, and emacs-lucid.  The first includes things that
>>> are needed by each of the other four, and those four each include an
>>> emacs binary.  The binary in the emacs package is
>>> /usr/bin/emacs-nox.exe.  The other packages contain
>>> /usr/bin/emacs-X11.exe, and so on.
>>>
>>> This way of naming the packages doesn't really reflect the contents of
>>> the emacs package.  It also means that anyone who installs emacs gets
>>> emacs-nox.exe, even if they plan to use one of the other three
>>> binaries.
>>>
>>> I would rather rename the current emacs-common package to emacs and
>>> the current emacs package to emacs-nox.  But then the new emacs would
>>> have to have a way of requiring the installation of at least one of
>>> emacs-nox, emacs-X11, emacs-w32, or emacs-lucid.  Is there any way to
>>> do this with our current setup machinery?
>>
>> I don't think we have the transaction capability that would be necessary
>> to specify that the meaning of an already existing package name (two,
>> actually) changes in this manner.  You might have to use new package
>> names and place the appropriate obsoletions w.r.t. old names instead.
>>
>>> My idea three years ago was to have the new emacs package require a
>>> "feature" called, for instance, emacs-bin, and then have each of
>>> emacs-nox, emacs-X11, emacs-w32, emacs-lucid "provide" that feature.
>>
>> I have no idea if multiple packages can provide the same feature. It's
>> worth trying if it does what you want (you must pick at least one of
>> those).
> 
> This seems to work, with one caveat.  Suppose package P requires feature 
> f, and packages Q, R, S,... provide f.  If the user selects P and one or 
> more of Q, R, S,..., setup is happy.  But if the user simply selects P, 
> then setup/libsolv will choose among Q, R, S,... the one whose name is 
> alphabetically first.  In the emacs case, this would be emacs-lucid, 
> which is a stupid default.  The default ought to be emacs-nox.  So I can 
> make it work if I call that package emacs-basic instead of emacs-nox.

Yeah, I think what's wanted here is for the solver to output a problem 
with the choices, rather than picking one.  I'm not sure how to get it 
to do that.

(Ofc, then we need some UI for picking problem solutions, rather than 
just always using the default)

(and I'm not sure how we'd encode "emacs-basic" should be the default 
provider of "emacs-bin" as the input into the solver; presumably there'd 
by some scheme with weights attached to provide names to set the order 
rather than alphabetic)

> I've only tested setup so far, not calm.  Jon, if you're reading this, 
> does calm allow 'requires' and 'provides' to contain arbitrary names 
> that are not package names?

Yes.