Question

BiocGenerics built with R 3.4.2 changes behaviour

4

Entering edit mode

ginggs ▴ 40

@ginggs-14076

Last seen 6.6 years ago

Building BiocGenerics with the admittedly very new R 3.4.2 causes failures in unit tests of at least IRanges and VariantAnnotation.

IRanges RUnit Tests - 91 test functions, 2 errors, 0 failures
ERROR in test_AtomicList_general: Error in match.arg(method) : 'arg' must be of length 1
ERROR in test_AtomicList_numerical: Error in match.arg(method) : 'arg' must be of length 1

VariantAnnotation RUnit Tests - 80 test functions, 1 error, 0 failures
ERROR in test_VRanges_vcf: Error in match.arg(method) : 'arg' must be of length 1

Any suggestions on how to troubleshoot this further?

biocgenerics iranges variantannotation • 2.1k views

ADD COMMENT • link updated 6.6 years ago by Hervé Pagès 16k • written 6.6 years ago by ginggs ▴ 40

2

Entering edit mode

Hi,

After installing R 3.4.2 I was able to reproduce this. The upgrade to R 3.4.2 also caused many other failures in many other Bioconductor packages. We advice users to NOT upgrade their installation to R 3.4.2 while we investigate this.

Cheers,

H.

ADD REPLY • link 6.6 years ago Hervé Pagès 16k

0

Entering edit mode

Thanks for confirming! Sorry, I didn't see your reply until I had already sent the follow up comment.

ADD REPLY • link 6.6 years ago ginggs ▴ 40

score 1 · Answer 1 · 2017-10-06

There is a regression in R 3.4.2 that affects S4 generics with dispatch on the ellipsis. The bug is in the methods package and breaks code like sapply(x, paste, collapse="") (where x is a list of character vectors) by not passing the collapse argument to the internal calls to paste(). We've implemented a workaround in BiocGenerics that should fix generic functions paste, order, cbind, rbind, mapply, pmax, pmin, pmax.int, and pmin.int.

The fix is in BiocGenerics 0.22.1 (release) and 0.23.3 (devel). Both versions should become available via biocLite() in the next 12-15 hours.

Cheers,

H.

score 0 · Answer 2 · 2017-10-04

This is already partially fixed in git, S4Vectors:

Work around regression in R 3.4.2 that breaks "order" (and indirectly "sort") method for List objects.
More work around regression in R 3.4.2 that breaks "order" (and indirectly "sort") method for List objects.

and IRanges:

Fix default value for 'method' argument in "order" method for Ranges objects (must be the same as in base::order()).