Re: [ANN] org-ql 0.4 released

[Michael Alan Dorman <mdorman@jaunder.io>]

Adam,

I do recognize that melpa-stable is not in any practical way curated,
guaranteed, cross-tested, etc.---and that emacs' packaging doesn't even
necessarily provide what might be required for anyone to try to do any
of that.

On the other hand, I *also* don't assume that maintainers are incapable
of making a reasonable assessment of the stability of their packages, or
of making a personal choice to try to maintain API compatibility in some
sensible way, and so forth.

And you know what: my personal experience over the last five years
hasn't been subject to the problems you identify; perhaps I'm just
lucky.

Nevertheless, my experience leads me to be of the opinion that
abolishing melpa-stable would reek of making the perfect the enemy of
the good---breakage can happen, and some things on it are no better than
whatever happens to be up on melpa, and it's unlikely to ever improve to
be a perfectly curated set of package versions, and got knows some stuff
lags behind enormously...but I think that it is still an improvement to
give maintainers *some* strategy for trying to manage their packages and
their dependencies and communicate all this to their users, rather than
consigning every emacs user to doing individual curation for every
single package they ever use.

Given your position, though, could I suggest that you at least remove
dependencies from your packgaes that feature versions that can only make
sense with melpa-stable?  That's what ultimately started this: the fact
that your new release of org-ql depends on a version of org-super-agenda
that *looks* like you care about melpa stable.

Mike.

Adam Porter <adam@alphapapa.net> writes:
> Michael Alan Dorman <mdorman@jaunder.io> writes:
>
>>> Hi friends,
>>>
>>> FYI, I've released org-ql 0.4.  It includes many improvements since 0.3.
>>>
>>> https://github.com/alphapapa/org-ql
>>
>> It would be nice if you could do a stable release of org-super-agenda so
>> that it could be installed from melpa-stable...
>
> Comments like yours lead me to the conclusion that MELPA Stable needs to
> be abolished.  I have been a proponent of the idea of MELPA Stable, so I
> don't say that lightly.
>
> I'll assume that you don't know what the technical issues are and offer
> an explanation.  Briefly:
>
> + MELPA Stable is nothing like what one might assume it's intended to be
>   like, e.g. Debian Stable or Debian Testing.  It provides none of the
>   benefits that Debian Stable and Testing provide.
>
> + MELPA Stable is, just like "regular" MELPA, a "rolling" collection of
>   packages developed without any coordination between maintainers.
>
> + The only difference is that MELPA Stable contains whatever versions of
>   packages that their maintainers have decided to tag with a version
>   number, rather than the latest commit to the master branch.  These
>   versions are not necessarily better, more stable, more reliable, or
>   more trustworthy than the untagged versions which appear in "regular"
>   MELPA.
>
> + Due to the lack of coordination between dependencies and their APIs,
>   version conflicts and breakage are a regular occurrence.  For example,
>   if package A depends on package B, and package B makes an API change
>   and tags a new MELPA Stable release, users of package A's MELPA Stable
>   version will see package A cease to work properly until package A, not
>   only commits a fix, but tags a new MELPA Stable version containing the
>   fix.  Since packages A and B do not share the same development
>   schedule, it is likely that their tagged-version release schedules
>   will not synchronize well.
>
>   If you are familiar with Debian, imagine if any upstream changes were
>   automatically pushed to Testing despite any freeze that might be in
>   place.  It would be virtually impossible to complete a freeze and make
>   a new stable release, and Testing and Stable would cease to be useful,
>   leaving only Unstable as a usable target.  This is the situation
>   between "regular" MELPA and MELPA Stable.
>
> For my packages, I tag stable versions for a few reasons:
>
> + To help users track changes in the changelog.
>
> + To help me separate new, possibly bug-introducing changes from
>   working, debugged code.
>
> + To help packagers in systems like Debian and Guix, who package stable
>   versions of some Elisp packages--and who, in so doing, take
>   responsibility for breakage.
>
> Now, I sympathize with not wanting to be vulnerable to potential
> breakage caused by the uncoordinated release of changes to packages on
> "regular" MELPA.  That is a real problem.  But the solution is not to
> use MELPA Stable.  The solution is to take charge of what packages you
> upgrade and when, rather than being at the mercy of whatever commits
> happen to be in MELPA at the moment.
>
> For myself, I commit the ~/.emacs.d/elpa directory to Git with the rest
> of my config, and I upgrade packages intentionally.  If breakage
> happens, I can easily revert and deal with it later.  Other users use
> alternative package managers, like Borg or Straight or Quelpa, which
> pull changes directly from Git repos and allow pinning to commits, tags,
> etc.
>
> So, for yourself, I can only recommend that you abandon MELPA Stable and
> install packages by other means.  If you don't have the time or
> inclination to redo your whole config like that, then just use something
> like Quelpa to install the current version of org-super-agenda directly.
> It's a couple lines of use-package in your config, and you can upgrade
> it manually from then on, e.g. with
> <https://github.com/alphapapa/unpackaged.el#upgrade-a-quelpa-use-package-forms-package>.
> As always, your Emacs config is up to you.
>
> Now, I'm off to to the discussions on MELPA's tracker to add my vote to
> abolish MELPA Stable, or to at least allow packages to opt-out of it.