Import of BigWig files with rtracklayer fails under MultiCoreParam

Start by running the reproducible example with bplapply, comparing SerialParam and MulticoreParam around rtracklayer's import and empty BigWig tiles. Trace the interaction between MulticoreParam workers and import, then confirm the cause and define a regression check covering empty and non-empty regions.

I'm seeing a weird bug in in my CAGEfightR package when trying to import BigWigFiles.

Here's a reproducible example:

# Load the example data
library(BiocParallel)
library(rtracklayer)
library(CAGEfightR)
data('exampleDesign')

# Example bigwig files
bw_plus <- system.file('extdata', exampleDesign$BigWigPlus,
                       package = 'CAGEfightR')
bw_plus <- BigWigFileList(bw_plus)

# Genome
mm9 <- SeqinfoForUCSCGenome("mm9")
grl <- tileGenome(mm9, ntile=3)

# Import, knowing that some areas will be empty
import(bw_plus[[1]], which=grl[[1]]) # empty
import(bw_plus[[1]], which=grl[[2]]) # empty
import(bw_plus[[1]], which=grl[[3]]) # has signal

# This works
lapply(bw_plus, import, which=grl[[1]])
lapply(bw_plus, import, which=grl[[2]])
lapply(bw_plus, import, which=grl[[3]])

# This works
register(SerialParam())
bplapply(bw_plus, import, which=grl[[1]])
bplapply(bw_plus, import, which=grl[[2]])
bplapply(bw_plus, import, which=grl[[3]])

# This works
register(MulticoreParam(3))
bplapply(bw_plus, import)

# This fails!
register(MulticoreParam(3))
bplapply(bw_plus, import, which=grl[[1]])
bplapply(bw_plus, import, which=grl[[2]])
bplapply(bw_plus, import, which=grl[[3]])

To summarize:

• You can import entire BigWig-files using bplapply.
• You can import part of a BigWig using bplapply and SerialParam
• You cannot import an empty tile from a BigWig file with bplapply and MultiCoreParam.

What's causing this strange behavior?