Hi all,

I stumbled upon a weirdness caused by loading of package S4Vectors. which seems to break the ifelse function. Here's the example code (maybe not the simplest, but hopefully simple enough):

################ Code begin

testFnc = function(exprData, colors, universalColors = NULL,
           grey = ifelse(is.null(universalColors), colors, "grey"))
{
  print(is.null(universalColors));
  print( grey);
}

expr1 = 1;
labels1 = 1

MEs1 = testFnc(expr1, universalColors = labels1)

library(S4Vectors)
MEs1 = testFnc(expr1, universalColors = labels1)

################# Code end

The first call to testFnc produces the expected result:

[1] FALSE
[1] "grey"

The second call to testFnc, after loading S4Vectors, produces the following output:

[1] FALSE
Error in ifelse(is.null(universalColors), colors, "grey") :
  error in evaluating the argument 'yes' in selecting a method for function 'ifelse': Error: argument "colors" is missing, with no default

This means that even though the 'test' in the ifelse is FALSE, the yes argument is evaluated. This breaks one my WGCNA functions... thanks in advance for looking into it!

Peter

> sessionInfo()
R version 3.2.2 beta (2015-08-05 r68859)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Fedora 20 (Heisenbug)

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C            
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8  
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8  
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                
 [9] LC_ADDRESS=C               LC_TELEPHONE=C          
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C      

attached base packages:
[1] parallel  stats4    stats     graphics  grDevices utils     datasets
[8] methods   base    

other attached packages:
[1] S4Vectors_0.6.3     BiocGenerics_0.14.0

loaded via a namespace (and not attached):
[1] tools_3.2.2

Hi Peter,

All S4Vectors is doing is turning ifelse() into a generic function with dispatch on its 3 arguments. As a result, the arguments passed to it get evaluated before dispatch can happen, like for any generic function. I understand this changes a little bit the contract of the ifelse() function. According to its man page:

     ‘yes’ will be evaluated if and only if any element of ‘test’ is true, and analogously for ‘no’.

One workaround is for you to use if else instead. See this recommendation from Martin Maechler:

    https://stat.ethz.ch/pipermail/r-help/2014-December/424367.html

If this is too much of an annoyance we could also get rid of the ifelse methods for Rle objects defined in S4Vectors:

> showMethods("ifelse")
Function: ifelse (package base)
test="ANY", yes="ANY", no="ANY"
test="ANY", yes="ANY", no="Rle"
test="ANY", yes="Rle", no="ANY"
test="ANY", yes="Rle", no="Rle"

These methods were added a long time ago when Rle objects were implemented in IRanges. I moved them from IRanges to S4Vectors when I did the IRanges/S4Vectors split. I don't know how much use they get.

This is a little bit extreme though. What other alternatives are there?

Cheers,

H.

After 2 years it is still broken. This time IRanges introduce faulty code.

a <- "foo"
ifelse(is.numeric(a), yes=a+5, no=a)

[1] "foo"

library(IRanges)
ifelse(is.numeric(a), yes=a+5, no=a)
Error in a + 5 : non-numeric argument to binary operator