gcc undefined behavior

[ilikegitlab]

I am hitting a CRAN undefined behavior check. The simplest way to reproduce is:

library(units)
vec <- c(1)
set_units(1, "1") - set_units(vec[F], "%")

normally this runs fine, but within the CRAN "additional issues" gcc-UBSAN environment:

/usr/local/lib/R/site-library/Rcpp/include/Rcpp/vector/traits.h:49:62: runtime error: reference binding to misaligned address 0x000000000001 for type 'type', which requires 8 byte alignment
0x000000000001: note: pointer points here
<memory cannot be printed>
/usr/local/lib/R/site-library/Rcpp/include/Rcpp/vector/Vector.h:340:62: runtime error: reference binding to misaligned address 0x000000000001 for type 'type', which requires 8 byte alignment
0x000000000001: note: pointer points here
<memory cannot be printed>
/usr/local/lib/R/site-library/Rcpp/include/Rcpp/vector/traits.h:49:62: runtime error: reference binding to misaligned address 0x000000000001 for type 'type', which requires 8 byte alignment
0x000000000001: note: pointer points here
<memory cannot be printed>
/usr/local/lib/R/site-library/Rcpp/include/Rcpp/vector/Vector.h:340:62: runtime error: reference binding to misaligned address 0x000000000001 for type 'type', which requires 8 byte alignment
0x000000000001: note: pointer points here
<memory cannot be printed>
 [1]
Warning messages:
1: In ud_convert(unclass(x), str1, str2) :
  subscript out of bounds (index 0 >= vector size 0)
2: In ud_convert(unclass(x), str1, str2) :
  subscript out of bounds (index 0 >= vector size 0)

Now I agree its a bit strange code (and I can work around this), but sometimes operations on large vectors are only done on some elements, and sometimes there are no elements left in the vector. So perhaps it is something that can or should be addressed in ud_convert (as the UB is caused by the Rcpp glue).

[Enchufa2]

Certainly, thanks.