The most likely culprit is C code that is modifying a logical vector
without checking whether this is legitimate for R semantics
(i.e. making sure MAYBE_REFERENCED or at least MAYBE_SHARED is FALSE).
If that is the case, then this is legitimate for C code to do in
principle, so UBSAN and valgrind won't help. You need to set a gdb
watchpoint on the location, catch where it is modified, and look up
the call stack from there.
The error signaled in the GC is a sanity check for catching that this
sort of misbehavior has happened in C code. But it is a check after
the fact; it can't tell you more that that the problem happened
sometime before it was detected.
Best,
luke
On Wed, 13 Oct 2021, Martin Morgan wrote:
The problem with using gdb is you'd find yourself in the garbage collector, but
perhaps quite removed from where the corruption occurred, e.g., gc() might /
will likely be triggered after you've returned to the top-level evaluation
loop, and the part of your code that did the corruption might be off the stack.
The problem with devtools::check() (and R CMD check) is that running the unit
tests occurs in a separate process, so things like setting a global option (and
even system variable from within R) may not be visible in the process doing the
check. Conversely, for the same reasons, it seems like the problem can be
tickled by running the tests alone. So
R -f <your_source>/tests/testthat.R
would seem to be a good enough starting point.
Actually, I liked Henrik's UBSAN suggestion, which requires the least amount of
work. I think I'd then try
R -d valgrind -f <your_source>/tests/testthat.R
and then further into the weeds... actually from the section of R-exts you
mention
R_C_BOUNDS_CHECK=yes R -f <your_source>/tests/testthat.R
might also be promising.
Martin
On 10/12/21, 10:30 PM, "Bioc-devel on behalf of Pariksheet Nanda"
<bioc-devel-boun...@r-project.org on behalf of pariksheet.na...@uconn.edu> wrote:
Hi all,
On 10/12/21 6:43 PM, Pariksheet Nanda wrote:
>
> Error in `...`: internal logical NA value has been modified
In the R source code, this error is in src/main/memory.c so I was
thinking one way of investigating might be to run `R --debugger gdb`,
then running R to load the symbols and either:
1) set a breakpoint for when it reaches that particular line in
memory.c:R_gc_internal and then walk up the stack,
2) or set a watch point on memory.c:R_gc_internal:R_LogicalNAValue
(somehow; having trouble getting gdb to reach that context).
3) Then I thought, maybe this is getting far into the weeds and instead
I could check the most common C related error by enabling bounds
checking of my C arrays per section 4.4 of the R-exts manual:
$ R -q
> options(CBoundsCheck = TRUE)
> Sys.setenv(R_C_BOUNDS_CHECK = "yes") # Try both ways *shrug*
> devtools::test()
... # All tests still pass.
> devtools::check()
... # No change :(
Maybe I'm not sure I'm using that option correctly? Or the option is
ignored in devtools::check(). Or indeed, the error is not from over
running C array boundaries.
It turns out that using the precompiled debug symbols[1] isn't all that
useful here because I don't get line numbers in gdb without the source
files and many symbols are optimized out, so it looks like I would need
to compile R from source with -ggdb first instead of using the Debian
packages.
Hopefully this is still the right approach?
Pariksheet
[1] After install r-base-core-dbg on Debian for the debug symbols.
_______________________________________________
Bioc-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/bioc-devel
_______________________________________________
Bioc-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/bioc-devel
--
Luke Tierney
Ralph E. Wareham Professor of Mathematical Sciences
University of Iowa Phone: 319-335-3386
Department of Statistics and Fax: 319-335-3017
Actuarial Science
241 Schaeffer Hall email: luke-tier...@uiowa.edu
Iowa City, IA 52242 WWW: http://www.stat.uiowa.edu
_______________________________________________
Bioc-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/bioc-devel
