Reproducible example for question on Bioc-Devel https://stat.ethz.ch/pipermail/bioc-devel/2014-August/006106.html

extraColumnSlots_confusion.R

## Necessary packages (BioC-devel)
library(GenomicRanges)
library(S4Vectors)

## Class A using a DataFrameOrNULL in internalPos slot
setClassUnion(name = "DataFrameOrNULL", members = c("DataFrame", "NULL"))

setClass("A",
         contains = "GRanges",
         representation(
           internalPos = "DataFrameOrNULL", 
           size = "integer"),
         prototype(
           internalPos = NULL,
           size = NA_integer_)
)

setMethod(GenomicRanges:::extraColumnSlotNames, "A",
          function(x) {
            c("internalPos")
          })

A <- function(seqnames = Rle(), tuples = matrix(), 
              strand = Rle("*", length(seqnames)), ..., 
              seqlengths = NULL, seqinfo = NULL) {
  
  # Get size of tuples
  if (all(is.na(tuples))) {
    size <- NA_integer_
  } else {
    size <- ncol(tuples)
  }
  
  # Create IRanges
  if (!is.na(size)) {
    ranges <- IRanges(start = tuples[, 1], end = tuples[, size])
  } else {
    ranges <- IRanges()
  }
  
  # Create internalPos
  if (is.na(size) || size < 3) {
    internalPos <- NULL
  } else {
    internalPos <- DataFrame(tuples[, seq(from = 2L, to = size - 1, by = 1L), 
                                    drop = FALSE])
  }
  
  # Create GRanges
  gr <- GRanges(seqnames = seqnames, ranges = ranges, strand = strand, 
                seqlengths = seqlengths, seqinfo = seqinfo, ...)
  
  new("A", gr, internalPos = internalPos, size = size)
}

# Class B using matrixOrNULL in internalPos slot
setClassUnion(name = "matrixOrNULL", members = c("matrix", "NULL"))
setClass("B",
         contains = "GRanges",
         representation(
           internalPos = "matrixOrNULL", 
           size = "integer"),
         prototype(
           internalPos = NULL,
           size = NA_integer_)
)

setMethod(GenomicRanges:::extraColumnSlotNames, "B",
          function(x) {
            c("internalPos")
          })

B <- function(seqnames = Rle(), tuples = matrix(), 
              strand = Rle("*", length(seqnames)), ..., 
              seqlengths = NULL, seqinfo = NULL) {
  
  # Get size of tuples
  if (all(is.na(tuples))) {
    size <- NA_integer_
  } else {
    size <- ncol(tuples)
  }
  
  # Create IRanges
  if (!is.na(size)) {
    ranges <- IRanges(start = tuples[, 1], end = tuples[, size])
  } else {
    ranges <- IRanges()
  }
  
  # Create internalPos
  if (is.na(size) || size < 3) {
    internalPos <- NULL
  } else {
    internalPos <- tuples[, seq(from = 2L, to = size - 1, by = 1L), 
                          drop = FALSE]
  }
  
  # Create GRanges
  gr <- GRanges(seqnames = seqnames, ranges = ranges, strand = strand, 
                seqlengths = seqlengths, seqinfo = seqinfo, ...)
  
  new("B", gr, internalPos = internalPos, size = size)
}

# Function to make two tuples object of class A or B of given size and replace 
# an element of the first with an element of the second to create a third 
# object. Essentially, make a call to S4Vectors::replaceROWS with each class.
# If size < 3 then internalPos is NULL otherwise it is a DataFrame 
# (resp. matrix) with "size - 2" columns.
f <- function(size = 3, class = c("A", "B")) {
  
  seqinfo <- Seqinfo(paste0("chr", 1:3), c(1000, 2000, 1500), NA, "mock1")
  
  if (class == "A") {
    # Create object of class A
    x <- A(seqnames = Rle('chr1', 10), 
           tuples = matrix(seq(1, size * 10), ncol = size), seqinfo = seqinfo)
    # Create another object of class A
    value <- A(seqnames = Rle('chr3', 1), 
            tuples = matrix(seq(size * 10 + 1, size * 10 + size), ncol = size),
            seqinfo = seqinfo)
    # Try to replace the first element of a with aa
    xx <- replaceROWS(x, 1, value)
  } else if (class == "B") {
    # Create object of class B
    x <- B(seqnames = Rle('chr1', 1), 
           tuples = matrix(seq(1, size * 10), ncol = size), seqinfo = seqinfo)
    # Create another object of class A
    value <- B(seqnames = Rle('chr3', 1), 
            tuples = matrix(seq(size * 10 + 1, size * 10 + size), ncol = size), 
            seqinfo = seqinfo)
    # Try to replace the first element of a with aa
    xx <- replaceROWS(x, 1, value)
  }
  
  return(xx)
}

## Define a replaceROWS function with signature NULL
## Perhaps more generally useful than my package?
setMethod("replaceROWS", 
          "NULL",
          function(x, i, value) {
            NULL
          }
)

## Class A - works regardless of size. By this I mean replaceROWS also updates 
# the internalPos slot.
# NB: the show method (which is inherited from GenomicRanges) will fail when 
# size < 3 but this isn't important here. Use str() to see that the object 
# was indeed properly updated.
a1 <- f(1, "A")
a1@internalPos
a2 <- f(2, "A")
a2@internalPos
# Works if size > 2. By this I mean it also updates the internalPos slot.
a3 <- f(3, "A")
a3@internalPos
a4 <- f(4, "A")
a4@internalPos

## Class B - works if size < 3, fails otherwise.
# NB: the show method (which is inherited from GenomicRanges) will fail when 
# size < 3 but this isn't important here. Use str() to see that the object 
# was indeed properly updated.
b1 <- f(1, "B")
b1@internalPos
b2 <- f(2, "B")
b2@internalPos
# Fails if size = 3 due to "ans_ecs <- GenomicRanges:::extraColumnSlotsAsDF(x)" 
# in the replaceROWS method for B (which is inherited from GenomicRanges).
# This means that the subsequent call to 
# "update(x, seqnames = ans_seqnames, ranges = ans_ranges, strand = ans_strand, 
# elementMetadata = ans_mcols, .slotList = as.list(ans_ecs))" in replaceROWS  
# gets a DataFrame, rather than a matrix, for the internalPos slot and 
# causes the error.
b3 <- f(3, "B")
# Fails with a slightly different error if size > 3, which is again due to 
# "ans_ecs <- GenomicRanges:::extraColumnSlotsAsDF(x)" in the replaceROWS 
# method for B (which is inherited from GenomicRanges). The difference being 
# that it fails at the subsequent call to 
# "ans_ecs <- replaceROWS(ans_ecs, i, value_ecs[seq_len(ans_necs)])" in 
# replaceROWS
b4 <- f(4, "B")

## What I could do:
# (1): Just use class A, i.e. a DataFrame for internalPos, since this works 
# out-of-the-box.
# (2) : I could define a replaceROWS method for class B but it would be nice 
# for it to work via inheritance to GenomicRanges when the subclass has a 
# matrix in a slot that is an extraColumnSlots (unless this is not allowed).

sessionInfo()
# 
# R version 3.1.1 (2014-07-10)
# Platform: x86_64-apple-darwin13.1.0 (64-bit)
# 
# locale:
#   [1] en_AU.UTF-8/en_AU.UTF-8/en_AU.UTF-8/C/en_AU.UTF-8/en_AU.UTF-8
# 
# attached base packages:
#   [1] parallel  stats     graphics  grDevices utils     datasets  methods  
# [8] base     
# 
# other attached packages:
#   [1] GenomicRanges_1.17.35 GenomeInfoDb_1.1.18   IRanges_1.99.24      
# [4] S4Vectors_0.1.2       BiocGenerics_0.11.4  
# 
# loaded via a namespace (and not attached):
#   [1] packrat_0.4.0.12 stats4_3.1.1     tools_3.1.1      XVector_0.5.7