ChangeLog warning is nonsense

[eddelbuettel]

When checking a package in RStudio, devtools is used even though I might prefer R CMD check --as-cran.

Worse, devtools appears to be drunk as it warn about

   Non-standard files found:
     ChangeLog

A casual check on the CRAN site reveals that ~ 10% of all packages have such a file:

 edd@xmcorsairs:~/CRAN$ ls -1 */ChangeLog  | wc -l
 555
 edd@xmcorsairs:~/CRAN$ 

Can we please turn that warning off?

And I'd say ditto for the warning about configure.in though this is a different issue. Why be holier than the Pope? If R CMD check does not warn...

[jjallaire]

FYI you can disable the use of devtools by going to the package build options.

[eddelbuettel]

Yes and I now have but I

• dislike having to do that over and over for each project on each machine
• devtools is still wrong, whether I choose to ignore it or not

[eddelbuettel]

And, of course CRAN is perfectly happy to display a ChangeLog file alongside a NEWS and README file -- eg Rcpp has all three, as do a few others of my packages.

[jjallaire]

I think we could add a global preference to RStudio that controls whether new projects get this by default (so you could set this once per machine for all time and be done with it)

[eddelbuettel]

Yes, I would welcome that.

I would also welcome a fix to the 'ChangeLog does not belong here' issue. At best, you are imposing personal preferences. The file is a GNU coding standard. You can do better than this.

[jjallaire]

Okay, I just added a global preference (under Packages) that you can flip to ensure that new packages always use plain R CMD check. I am sure it will be also straightforward to modify devtools to not warn for ChangeLog.

[eddelbuettel]

Thank you. We can leave the devtools change to its strong-willed maintainer.

Configure.in also belongs in a tarball if I as maintainer decide it belongs. It governs / documents the build and is source code.

Overall, one cannot simultaneously complain about CRAN Policy being inconsistent and poorly communicated and then make up one's own checks out of this air like this. I am really surprised by this, and not in positive way...

[hadley]

1. As you are aware, CRAN maintainers sometimes require things that are not tested by for R CMD check. When this happens we add our own checks - this is why check() does more than just run R CMD check --as-cran.
2. Sometimes our own checks add false positives. False positives are a part of automated checking.
3. "Never attribute to malice that which can be adequately explained by stupidity."
4. We will fix the bug, and we're happy to review a pull request.

[eddelbuettel]

Thanks for the feedback. I'd be happy to help, time permitting.

Re 1), we'd need a clear pulished list to go by. Otherwise locally extending what is guestimated to be R CMD check's intent is madness. I am all for checking leftover / backup files from editors, IDEs etc but actual working source is where I draw the line.

[wch]

I'm not sure, but it might be possible to remove the devtools non-standard file check. I noticed recently that R CMD check started giving notes or warnings for non-standard files. (I'm not sure if this was with R 3.0.3 or with R-devel.)