decisionpatterns / formula.tools Goto Github PK

Fix a dependency on testthat which caused CRAN tests to fail.

A number of methods are defined for the "<-" class.

This breaks CRAN checks.

The rhs.vars function cannot identify correctly the variables whenever there is a minus sign on the right-hand side of the formula.

Example:

f1 <- x ~ a + b - c
rhs(f1) # Ok
> rhs(f1)
[1] a + b - c
rhs.vars(f1) # c is missing
> rhs.vars(f1)
[1] "a" "b"

I'm ran this code in a Windows machine, formula.tools version 1.7.1.

The following code in which left-hand side contains a function (e.g., log(Girth)) returns a list while character vector is expected:

rez <- lhs.vars(log(Girth) ~ Volume, data = trees)
class(rez)

[1] "terms"   "formula"

In the same situation get.vars() returns a list that is also either unexpected (or undocumented).

Is this behavior caused by a bug?

Hi,

Love the package, thanks for sharing it.

I'm having problems with get.vars with signature NULL. This comes up in a situation like this

 lhs.vars( ~ x)

which then calls get.vars on lhs(~x). Since lhs(~x) returns NULL this leads to get.vars being called with signature NULL, NULL and this throws an error.

In prior versions this would work and return NULL, which makes sense to me.

I tried to look in the NEWS to see if this was an expected change but there is no entry for 1.5.

Thanks for your time!

Best,
Paul

Loading the package breaks R's behaviour of as.character.default, which is not too convenient.

as.character(.~.) # "~" "." "."
library(formula.tools)
as.character(.~.) # ". ~ ."

Do you see an easy way to avoid this?

I'm not sure why this happens, but this throws an error only if the formula.tools package is loaded. I'm guessing it must be a bug in formula.tools:

library(formula.tools)
library(lmerTest)
m1 <- lmer(Informed.liking ~ Gender*Information +(1|Consumer), data=ham)
difflsmeans(m1, test.effs="Gender:Information")

It might be advantageous, long-term to refactor this package using the considerable less complex S3 Methods. Most of the functions do not need multiple dispatch. And this obviates some problems with CRAN/roxygen2 incompatibilities.

It would be nice to add this functionality into the package.

Perhaps through replace.call and replace.name functions

Using 1.5.4 I notice this bug when trying to assign lhs() in a one-sided formula:

f <- ~ A   
lhs(f)     
## NULL
lhs(f) <- quote(B)
f
## ~B

The expected result would be B ~ A, but instead the right-hand side gets replaced. There are no such issues when there is already something in the left-hand side:

f2 <- ZZZ ~ A   
lhs(f2) <- quote(B)
f2
## B ~ A

The name formula.tools does not accurately reflect the extent of the package functionality since the tools work on expressions and list as well. Most of this package should become expression.tools with the few formula specific functions remaining in this package.