Re: Sorting of '-' and '.' is equal in versions of R packages (Was: Do R autopkgtests not respect versioned Deps?)
On Tue, Sep 21, 2021 at 08:20:24AM -0500, Dirk Eddelbuettel wrote:
>
> Yes. CRAN, being more liberal, allows expressions that can lead to errors for
> us. I have always found these cases to be exceedingly rare. We have now ~1k
> r-{cran,bioc} packages and this comes up ... twice?
I gave three examples but stoped seeking for more. These examples cost
time - about 3-5 times compared to packages without this. I'm fine to
spent time on things if I can see some sense in it - but I fail to see
any sense here.
> | So I had to manually specify this in d/control, and that fixed the problem.
> | But curious, would you have a way to make both dh-R and the debian version agree to pull the right version?
>
> My apporach in those cases is (was) always to deal with the locally-caused
> issue locally (i.e. a patch if need, or a bit of code, and I may not have
> used dh-r at the time) rather than sending something upstream that is a
> non-issue for upstream. (I do not think we look good by spamming upstream
> for issues in our tests -- when those very tests pass at CRAN which is what
> upstream cares about -- but I made that point before and nobody listened.)
I was listening but I keep on disagreeing. If some CRAN package
maintainer decides to choose a '-' this is liberty. If some other CRAN
maintainer decides to refer to this version but replaces '-' by a '.' I
consider it a bug. The fact that two characters are sorting the same in
some system does not mean you can choose between those letters at random
- at least not if you want to avoid confusing users.
The fact that we have this discussion about once a year if not more
might be a good sign that there is room for enhancement.
Kind regards
Andreas.
--
http://fam-tille.de
Reply to: