Bug#868558: transition: r-api-3.4 [was Re: Bug#868558: nmu: multiple r-* packages]
On 28 September 2017 at 18:53, Sébastien Villemot wrote:
| On Thu, Sep 28, 2017 at 07:04:51AM -0500, Dirk Eddelbuettel wrote:
| >
| > On 28 September 2017 at 13:20, Sébastien Villemot wrote:
| > | On Thu, Sep 28, 2017 at 12:53:10PM +0200, Graham Inggs wrote:
| > | > On 28/09/2017 12:28, Sébastien Villemot wrote:
| > | > > Note that there are many arch:all R packages that will need sourceful upload
| > | > > (they are easy to identify on the transition tracker whose URL is above).
| > | >
| > | > Besides r-cran-nlp which doesn't show up in the tracker, I've found several
| > | > other arch:all packages that don't depend on r-api-3, but do pick up a
| > | > dependency on r-api-3.4 after a rebuild:
| > |
| > | This makes me wonder whether arch:all packages really need a dependency on r-api-*.
| > |
| > | If this value really tracks an API, as advertised, it makes sense. But if it
| > | actually tracks an ABI, as in the present case, then this situation is
| > | suboptimal and complicates transition.
| > |
| > | Maybe the best solution would therefore be to dissociate API and ABI tracking.
| > |
| > | Moreover, packages automatically pick up a versioned dependency on r-base-core.
| > | But this should not be necessary since we now have ABI tracking. It makes
| > | dependencies uselessly tight.
| > |
| > | Anyways, these (potential) improvements should probably wait for the next
| > | transition (planned in April if I understand correctly).
| >
| > There transitions, and then there are transitions. Let me explain:
| >
| > - right now a subset of 'source: any' package needs a rebuild; here we could
| > in fact discuss leaving 'source: all' out
| >
| > - R 3.5.0 will need a rebuild of all 'source: any' packages
| >
| > - In the past we rebuilds for nonAPI reasons: reorganisation of R's internal
| > help system (and internal file format) was one
| >
| > So we may as well through the big mantle of the so-called "API" transition
| > around all dependent packages. But we don't _have to_ right now.
| >
| > Can be argued either way. Do as you see fit.
|
| I now understand that we ideally need two API/ABI-like values instead of one:
|
| - one that is bumped when only arch:any packages need to be rebuilt
|
| - the other one that is bumped when both arch:any and arch:all packages should
| be rebuilt
|
| The first value would appear in the Depends of arch:any package, but not of
| arch:all ones; the second value would appear in the Depends of both arch:any
| and arch:all.
|
| Because, for this transition and for the next one (in April), we will have to
| make sourceful uploads of ~170 arch:all packages… that actually do not need to
| be rebuilt. And this is very painful because it must be done manually (contrary
| to rebuilds of arch:any packages that can be trigerred easily).
|
| If we adopted this scheme right now, that would save us a lot of work for the
| April transition (but not for this one, because we first have to convert
| arch:all packages to the new scheme).
|
| Please tell me what you think.
Well, sorry, but that is your baby now. I argued _this very issue of it being
different for R_ for two months or more, but nobody bought into it.
You all insisted on this approach which you now find more complicate, so here
it is. Your deal now.
(For what it is worth, and the R / Debian itersection in particular, the
RcppAPT (source) package allows you to query R's and Debian's package
meta-data. That was part of my analysis so I won't repeat it. Happy to help
if there are questions.)
I'll happily deal with / help with technical questions, I am not that
interested in managing a technical issue I argued (in vain) against for some
time.
Sorry, Dirk
--
http://dirk.eddelbuettel.com | @eddelbuettel | edd@debian.org
Reply to: