Warnings when reloading packages with S4 classes

[wch]

I get these warnings error when loading lubridate.

I think there are two issues.

• One is in unloadNamespace -> cacheMetaData -> .getGenerics -> .removeSuperclassBackRefs
• One is in unloadNamespace -> cacheMetaData -> .getGenerics -> .removeSuperclassBackRefs -> findClass, but I'm not yet sure where.

The first type of of warning only happens the second time a package is loaded with load_all(, TRUE). The second type of warning seems to happen every time except the first.

> load_all('lubridate', TRUE)
Loading lubridate

> load_all('lubridate', TRUE)
Loading lubridate
Warning messages:
1: In .removeSuperclassBackRefs(Class, classDef, classWhere) :
  could not find superclass "EnumerationValue" to clean up when removing subclass references to class "Duration"
2: In .removeSuperclassBackRefs(Class, classDef, classWhere) :
  could not find superclass "EnumerationValue" to clean up when removing subclass references to class "Interval"
3: In .removeSuperclassBackRefs(Class, classDef, classWhere) :
  could not find superclass "EnumerationValue" to clean up when removing subclass references to class "Period"
4: In FUN(X[[2L]], ...) :
  Created a package name, ‘2012-08-14 16:06:15’, when none found
5: In .removeSuperclassBackRefs(cl, cldef, searchWhere) :
  could not find superclass "EnumerationValue" to clean up when removing subclass references to class "Duration"
6: In .removeSuperclassBackRefs(cl, cldef, searchWhere) :
  could not find superclass "EnumerationValue" to clean up when removing subclass references to class "Interval"
7: In .removeSuperclassBackRefs(cl, cldef, searchWhere) :
  could not find superclass "EnumerationValue" to clean up when removing subclass references to class "Period"


> load_all('lubridate', TRUE)
Loading lubridate
Warning message:
In FUN(X[[2L]], ...) :
  Created a package name, ‘2012-08-14 16:07:25’, when none found

The package seems to work, though -- it passes all tests.

[wch]

Fixed and merged in 9330614. A second part needed fixing, at f57967a.

I changed it use importing code from loadNamespace. This also made it import packages and objects list in the NAMESPACE file, in the same way that they are loaded when using library().

Now the DESCRIPTION file isn't used to actually import packages, although it is still used to check version numbers.

[garrettgman]

Winston,

What was up here? I'm getting some of the behavior you describe above:

> library(devtools)
> devtools::test()

(the test output - everything passes)

Warning message:
In FUN(X[[2L]], ...) :
  Created a package name, ‘2013-12-17 11:06:29’, when none found

It's spooky that a date is involved, but I'm guessing that it's not a lubridate thing. Also, the warning message does not always appear, and it sometimes appears for load_all() -- even though the package loads fine.

[wch]

Oh yeah, I've seen this again recently. The date isn't a lubridate thing - I believe it has something to do with R's handing of .removeSuperclassBackRefs. Sorry I don't have more insight on it right now!

[wch]

Garrett - which package are you using when you see this?

With Shiny, this warning comes up now when you reload a package. It was introduced in rstudio/shiny@8d8ea53,

The problem is import(methods) in NAMESPACE. When that line is removed, reloading works fine.

[wch]

One more thing I just remembered: John Chambers suggests putting methods into the Depends field instead of importing it. It's the third message in this thread:
http://r.789695.n4.nabble.com/advise-on-Depends-td4678930.html

[garrettgman]

Winston,

I was using lubridate, of course :)

It's good to know that we can trace the problem to import(methods). Lubridate already lists methods in the dependency field. However, I add imports(methods) to pass R CMD check (it fails on a warning otherwise).

Is there an alternative way to use methods and pass CRAN's check?

[wch]

I'm thinking now that you can leave imports(methods) - the issue is with devtools. There's something different about how S4 classes are created when using load_all(), as compared to loading a built package with library(). Hopefully I'll be able to get load_all() to behave more like library().

[hadley]

Does this still happen? Or is it fixed in recent R?

[wch]

The "Created package name" warning seems to be gone on R 3.0.3, and R-devel (from about two weeks ago).

@garrettgman A heads up: on R-devel, I see these messages after the tests (but not on R 3.0.3):

Warning messages:
1: In Ops.factor(left, right) : - not meaningful for factors
2: In Ops.factor(left, right) : - not meaningful for factors
3: In Ops.factor(left, right) : - not meaningful for factors