[racket-dev] Packaging

I am saying we should use something that is not called "version
number".  On the IRC list I have suggested -- without too much thought
behind it yet -- that we construct an "upgrade graph"; package
maintainers can specify which package can be thought of as an
automatic improvement on another, and some appropriate part of the
Planet mechanism can therefore follow a chain of these links to find
the best available candidate for a require.  That allows package
names, version numbers, and other string-based user-readable
package-identifying features to be uninterpreted, and written however
the maintainer wants.

Carl Eastlund

On Tue, Feb 22, 2011 at 4:06 PM, Robby Findler
<robby at eecs.northwestern.edu> wrote:
> Carl: your message is unclear to me. Are you saying that attempting to
> solve the problem of matching up require requests with available
> versions of software packages is hopeless and we shouldn't attempt it,
> or are you saying that we should use something that is not (literally)
> called "version number" or are you saying something else?
>
> Robby
>
> On Tue, Feb 22, 2011 at 3:01 PM, Carl Eastlund <cce at ccs.neu.edu> wrote:
>> Do you mean to inherit Planet's current "version number" semantics?
>> Ugggghhhhh.  Assigning a fixed structure and semantics to version
>> numbers was one of the worst things Planet did.  Dracula is up to
>> 8:18, and goodness knows what that means.  It does not mean there have
>> been 8 significantly different versions of Dracula, such that I gave
>> the release bigger fanfare than usual.  It means there were 8 times
>> when some potential incompatibility between releases occurred to me
>> between the time of package creation and the time of upload.  That
>> should not be how version numbers are determined.  It is not at all
>> clear to me that version numbers should serve as an automatic metric
>> of compatibility or upgrade-ability; let's either come up with a
>> metric that is more to the point, or stop trying so hard to enforce
>> things like "no compatibility regressions" that are often hard to
>> detect in the first place.
>>
>> Carl Eastlund
>>
>> On Tue, Feb 22, 2011 at 3:37 PM, Jay McCarthy <jay.mccarthy at gmail.com> wrote:
>>> I don't feel strongly about this and you seem to, so supposing we
>>> support any conflicting installations, it makes sense for Planet 2.0
>>> to have both major and minor versions.
>>>
>>> Jay
>>>
>>> 2011/2/19 Robby Findler <robby at eecs.northwestern.edu>:
>>>> On Sat, Feb 19, 2011 at 10:18 AM, Robby Findler
>>>> <robby at eecs.northwestern.edu> wrote:
>>>>> It looks to me like you there is relevant, important metadata that
>>>>> you're making someone fold into an implicit place instead of an
>>>>> explicit one.
>>>>>
>>>>> Will you have a convention for these? What if I decide to call mine
>>>>> "libgtk2.0" and someone else calls theirs "somepackage-2"? That
>>>>> doesn't seem good fo
>>>>
>>>> [ Sorry; got distracted here and forgot to come back. ]
>>>>
>>>> That doesn't seem good for users who are trying to find packages.
>>>> Especially if I were to call mine "2-somepackage" (you may think this
>>>> far fetched, but if you look you should find an example of this in our
>>>> current collection tree....)
>>>>
>>>> Robby