Re: math/R 4.3.1 (Re: pkgtools/revbump re-written to be easier to use

[Greg Troxel <gdt%lexort.com@localhost>]

Makoto Fujiwara <makoto%ki.nu@localhost> writes:

> Hi, I will be updating math/R from 4.2.0 to 4.3.1.
> I'm wondering if revbump on R-* packages are necessary.
>
> My simple guess comes from Perl update. When Perl is updated
> all the p5-* may be revbumped.  But this is because of those
> binary is located on perl version based directory. That's why
> all p5-* should be revbumped.
>
> But for R-*, it does not seem to have such a situation.
> If the R-* package is not binary linked, 
> (or even binary is linked, if ABI is not changed)
> we don't need revbump on that particular package.

There are two parts to this:

  If an R package built with 4.2.0 will definitely work with R 4.3.1
  installed instead, because the ABI is unchanged, then theoretically
  you don't need to bump.

  If the package built for R contains 4.2.0 in its name, then it really
  needs bumping to avoid confusion.  That doesn't seem to be the case.


The big questions are:

  1) Does R document if there was or was not an ABI change?  Do they have a
  history of being careful about this, and did they document changes
  when the ABI changed in the past?

  2) When you took an existing system, and did 'make replace' of R, did all
  the R packages you had installed continue to work?

If that's iffy, you should lean to revbump.

> Anyway, I've run new revbump 3.4 for math/R as wiz@ instruction,
> thanks for really simple operation necessary.
>
> But "wc commitlist" shows 399 packages, and I think this is the count
> of all the R-* packages. Recent math/R failure indicated 400 children
> had affected.
>
> My guess is that the among R-* packages, some have buildlink3.mk.
> And only those should be revbumped, correct ?
>
> If it is the case, my dirty hack will filter to pick only the packages
> which has buildlink3.mk:
>
> beebox-03@makoto 12:34:26/230730(..usr/pkgsrc-cvs)% rm /tmp/real; touch /tmp/real ; foreach i (`cat  commitlist`)
> ls $i/buildlink3.mk &&  echo  $i >> /tmp/real
> end
>
> This is too dirty showing not find lines, but /tmp/real
> has 22 package, and I believe my intention is there.
>
> Sorry,
> the another fact that I've updated math/R-gss 2.2.5,
> and
> pkg_info -qL R-gss 2.2.5, has the line
>   /usr/pkg/lib/R/library/gss/libs/gss.so
>
> (and my local tree has buildlink3.mk there, which is not committed),
> I guess some packages need blk3 file recently and that's why I've
> added blk3 there
> (which I don't remember which yet, and
>   grep -r R-gss/buildlink3.mk
> does not find any).

I suspect that you are right that there is no need to revbump packages
that just install R code and do not build anything.