What can cause CRAN to show package errors from another package?

1
votes

I've just pushed a new version of package userfriendlyscience (version 0.5-0) to CRAN. The CHECK results for Windows show errors. Viewing these shows that they seem to be caused by another package, specifically mgcv, or perhaps another package in the dependency tree desiring to use previously exported functions from this package.

I installed the newest R-devel and ran RCMD CHECK --as-cran again, but this still completes without any errors etc. As far as I can see, the package I submitted 'works'.

So I assume that indeed, the errors shows by CRAN reside in other packages. But, how can that be? Shouldn't CRAN block submission of packages with such errors, in other words, shouldn't it be impossible for a package that causes errors to exist on CRAN?

Does anybody have any suggestions as to how I can start figuring out what exactly goes wrong, so that I can fix it (if it turns out to be this package after all) or notify the authors of this is caused by another package?

[note: the previous version did cause an error, which has been resolved in 0.5-0]

[For anybody who want to verify the successful installation in Windows: instructions for installing the uploaded source version as well as the compiled windows version are here]

1
rpackagecran
could not find function "scaleReliability" in one of your unit tests does not indicate an error in another package. Whatever gave you that idea? If you look closely where version 0.4-1 fails it is in your tests. Version 0.5-0 apparently can't be installed on Windows. - Roland
No, that's true - and that error has been solved in 0.5-0. I meant that version - sorry, I should have been more explicit. I'll edit the question to address this ambiguity. - Matherion
OK. Then have you installed your package successfully on a Windows OS? - Roland
I have no idea what happens there and how that could be a Windows specific problem. mgcv is probably imported indirectly by one of your dependencies. However, exportPattern("^[[:alpha:]]+") is not a good practice. Try being more specific in your exports and see if that solves the issue. - Roland
You could try if you can reproduce the problem on win-builder.r-project.org. - Roland

1 Answers

1
votes

(answering myself this in case others run into this)

This issue resolved itself. After a while, CRAN no longer indicated the same error.

On the basis of this and experiences with another upload, I currently suspect that CRAN isn't 'ready' at any given point; rather, it continually builds packages. Therefore, it's possible that a package happens to run into some errors in a window where one package has already been updated, but a package depending on that package not yet. The errors resulting from the incompatibility then show up for your package.

The solution, assuming this interpretation is accurate, would then be to wait a few days to see whether the situation resolves itself. If not, I'd suggest trying to contact the maintainer of the package that throws the error to see whether they maybe know what's happening.