[Bioc-devel] Strange "internal logical NA value has been modified" error

rhub::check(platform = "linux-x86_64-rocker-gcc-san")
rhub::check(platform = "ubuntu-rchk")

It is from base R

  https://github.com/wch/r-source/blob/a984cc29b9b8d8821f8eb2a1081d9e0d1d4df56e/src/main/memory.c#L3214

and likely indicates memory corruption, not necessarily in the code that triggers the error (this is when the garbage collector is triggered...). Probably in *your* C code :) since it's the least tested. Probably writing out of bounds.

This could be quite tricky to debug. I'd try to get something close to a minimal reproducible example.

I'd try to take devtools out of the picture, maybe running the test/testhat.R script from the command line using Rscript, or worst case creating a shell package that adds minimal code and can be checked with R CMD build --no-build-vignettes / R CMD check.

You could try inserting gc() before / after the unit test; it might make it clear that the unit test isn't the problem. You could also try gctorture(TRUE); this will make your code run extremely painfully slowly, which puts a big premium on having a minimal reproducible example; you could put this near the code chunks that are causing problems.

You might have success running under valgrind, something like R -d valgrind -f minimal_script.R.

Hope those suggestions help!

Martin


?On 10/12/21, 6:43 PM, "Bioc-devel on behalf of Pariksheet Nanda" <bioc-devel-bounces at r-project.org on behalf of pariksheet.nanda at uconn.edu> wrote:

    Hi folks,

    I've been told to ask some of my more fun questions on this mailing list
    instead of Slack.  I'm climbing the ladder of submitting my first
    Bioconductor package (https://gitlab.com/coregenomics/tsshmm) and feel
    like there are gremlins that keep adding rungs to the top of the ladder.
      The latest head scratcher from running devtools::check() is a unit
    test for a  trivial 2 line function failing with this gem of an error:

     > test_check("tsshmm")

    ?? Failed tests
    ????????????????????????????????????????????????????????????????
    ?? Error (test-tss.R:11:5): replace_unstranded splits unstranded into +
    and - ??
    Error in `tryCatchOne(expr, names, parentenv, handlers[[1L]])`: internal
    logical NA value has been modified
    Backtrace:
          ?
       1. ??testthat::expect_equal(...) test-tss.R:11:4
       2. ? ??testthat::quasi_label(enquo(expected), expected.label, arg =
    "expected")
       3. ?   ??rlang::eval_bare(expr, quo_get_env(quo))
       4. ??GenomicRanges::GRanges(c("chr:100:+", "chr:100:-"))
       5.   ??methods::as(seqnames, "GRanges")
       6.     ??GenomicRanges:::asMethod(object)
       7.       ??GenomicRanges::GRanges(ans_seqnames, ans_ranges, ans_strand)
       8.         ??GenomicRanges:::new_GRanges(...)
       9.           ??S4Vectors:::normarg_mcols(mcols, Class, ans_len)
      10.             ??S4Vectors::make_zero_col_DFrame(x_len)
      11.               ??S4Vectors::new2("DFrame", nrows = nrow, check = FALSE)
      12.                 ??methods::new(...)
      13.                   ??methods::initialize(value, ...)
      14.                   ??methods::initialize(value, ...)
      15.                     ??methods::validObject(.Object)
      16.                       ??base::try(...)
      17.                         ??base::tryCatch(...)
      18.                           ??base:::tryCatchList(expr, classes,
    parentenv, handlers)
      19.                             ??base:::tryCatchOne(expr, names,
    parentenv, handlers[[1L]])
    [ FAIL 1 | WARN 0 | SKIP 0 | PASS 109 ]


    The full continuous integration log is here:
    https://gitlab.com/coregenomics/tsshmm/-/jobs/1673603868

    The function in question is:


    replace_unstranded <- function (gr) {
         idx <- strand(gr) == "*"
         if (length(idx) == 0L)
             return(gr)
         sort(c(
             gr[! idx],
             `strand<-`(gr[idx], value = "+"),
             `strand<-`(gr[idx], value = "-")))
    }


    Also online here:
    https://gitlab.com/coregenomics/tsshmm/-/blob/ef5e19a0e2f68fca93665bc417afbcfb6d437189/R/hmm.R#L170-178

    ... and the unit test is:


    test_that("replace_unstranded splits unstranded into + and -", {
         expect_equal(replace_unstranded(GRanges("chr:100")),
                      GRanges(c("chr:100:+", "chr:100:-")))
         expect_equal(replace_unstranded(GRanges(c("chr:100", "chr:200:+"))),
                      sort(GRanges(c("chr:100:+", "chr:100:-", "chr:200:+"))))
    })


    Also online here:
    https://gitlab.com/coregenomics/tsshmm/-/blob/ef5e19a0e2f68fca93665bc417afbcfb6d437189/tests/testthat/test-tss.R#L11-L12

    What's interesting is this is *not* reproducible by running
    devtools::test() but only devtools::check() so as far as I know there
    isn't a way to interactively debug this while devtools::check() is going on?

    Every few days I've seen on that "internal ... value has been modified"
    which prevents me from running nearly any R commands.  Originally I
    would restart R, but then I found I could clear that error by running
    gc().  No idea what causes it.  Maybe some S4 magic?

    Yes, I have downloaded the mailing lists for bioc-devel, r-devel,
    r-help, and r-package-devel and see no mention of "value has been
    modified" [1].

    Any help appreciated.

    Pariksheet



    [1] Mailing lists downloader:
    #!/bin/bash -x

    for url in
    https://stat.ethz.ch/pipermail/{bioc-devel,r-{devel,help,package-devel}}/
    do
         dir=$(basename $url)
         wget \
        --timestamping \
        --no-remove-listing \
        --recursive \
        --level 1 \
        --no-directories \
        --no-host-directories \
        --cut-dirs 2 \
        --directory-prefix "$dir" \
        --accept '*.txt.gz' \
        --relative \
        --no-parent \
        $url
    done