Skip to content

[R-pkg-devel] Submitting breaking changes to CRAN

5 messages · Ege Rubak, Duncan Murdoch

Original
Ege Rubak
# 
Dear all,

I'm seeking advice on how to submit a new package version with breaking
changes to CRAN. I will try to make this short:

1. spatstat (<= 1.65) had grow to be very large with extensive
examples, tests, and documentation.
2. CRAN asked us to reduce package size and check time.
3. We reorganized the package into a new umbrella package spatstat 2.0
which Depends/Imports several subpackages named spatstat.xxxx.
4. All subpackages are now on CRAN.
5. We submitted spatstat 2.0-1 which breaks 79 reverse dependencies
because they e.g. call functions that have been moved from spatstat to
spatstat.xxxx.
6. All maintainers have been warned over a period of months and offered
detailed help to adjust their package. Many have reported back that
they have a new version ready that will work with spatstat (>=2.0) and
are waiting to submit until it is on CRAN.
7. We received notification on 23 Feburary that "package spatstat_2.0-
1.tar.gz has been auto-processed. The auto-check found problems when
checking the first order strong reverse dependencies.
Please reply-all and explain: Is this expected or do you need to fix
anything in your package? If expected, have all maintainers of affected
packages been informed well in advance? Are there false positives in
our results?"
8. We replied to all on the same day, 23 Feb, that this was expected
and maintainers had been informed. Since then we have no news.

Any advice on how to cross the finish line and get spatstat 2.0-1 on
CRAN without putting too big a burden on the CRAN volunteers?

I can only come up with a long shot:

Ask package maintainers to submit their spatstat 2.0 compatible package
to CRAN with an additional line in DESCRIPTION:

Additional_repositories: https://spatstat.r-universe.dev

Since spatstat 2.0-1 is available from this repository they may pass
the incoming checks on CRAN, but my hopes are not too high.

If this was successful the reverse dependencies would be compatible
with spatstat 2.0 and on CRAN and so spatstat 2.0 would break nothing
and we could resubmit.

Best regards,
Ege
Duncan Murdoch
# 
It may be too late to do this now, but you could use the approach that 
devtools used when it was broken up:  The main package imports functions 
from the new spatstat.xxxx packages and exports them.  This way it could 
be done with no breaking changes.  Reverse dependencies could change to 
depend on spatstat.xxxx at their leisure.

Duncan Murdoch
On 11/03/2021 10:18 a.m., Ege Rubak wrote:
Ege Rubak
# 
Dear Duncan,

Thanks for taking the time to read my message and for the constructive
idea. You are right that it is a bit late for us to do this now. Given
that spatstat (<=1.65) exports >2,500 objects which are now spread
across sub-packages we would obviously have to make a script to help us
reexport the functions and make documentation containing links to the
real man page. This might be doable, but one big downside is that we
then don't use this occasion to move package dependencies from spatstat
to the relevant spatstat.xxxx. If the packages don't fail I'm afraid
that a lot of maintainers wont change anything, and their package will
depend on the entire ensemble of spatstat packages rather than just the
relevant sub-package(s). For our usual end users this would also mean
that when they open the help file of given function after attaching
spatstat it will just contain a huge list of links to the real help
file they are looking for which is a nuisance.

Unless we get other really good suggestions we will wait a bit more on
a reply from CRAN (we sent a gentle reminder a week ago) and hopefully
learn what we should do differently to get the last bit in place.

Best regards,
Ege
On Thu, 2021-03-11 at 14:56 -0500, Duncan Murdoch wrote:
Duncan Murdoch
# 
One part of your first message that I don't understand is the reluctance 
of downstream package writers to update to the spatstat.xxxx versions.

You said "Many have reported back that they have a new version ready 
that will work with spatstat (>=2.0) and are waiting to submit until it 
is on CRAN."

Why wouldn't those new versions work already?  I'm guessing there is 
some incompatibility between the new spatstat.xxxx packages and the old 
spatstat 1.65 from before the split.  Can you describe what it is? 
Maybe there's a way to make the new packages compatible with the old 
one, so there would be no excuse for the downstream packages not to 
update right away.

Duncan Murdoch
On 11/03/2021 7:39 p.m., Ege Rubak wrote:
3 days later
Ege Rubak
# 
Dear Duncan,

Thanks for spending time on this. You are right that there were some
incompatibility issues for people with old spatstat installed along
side spatstat.xxxx, and we might have been fix some of that, but it was
not completely clear to us. However, the problem is now solved as
spatstat 2.0-1 is on CRAN thanks to Uwe and the rest of the team. The
waiting time was explained by the fact that CRAN tackles packages with
many broken reverse dependencies separately and they had recently
worked on another case.

Best,
Ege
On Fri, 2021-03-12 at 09:46 -0500, Duncan Murdoch wrote: