[Rd] failed to assign RegisteredNativeSymbol for splitString

[Andrew Piskorski]

I saw a warning from R that I don't fully understand.  Here's one way
to reproduce it:

  $ /usr/local/pkg/R-3.2-branch-20160718/bin/R --version | head -n 3 
  R version 3.2.5 Patched (2016-05-05 r70929) -- "Very, Very Secure Dishes" 
  Copyright (C) 2016 The R Foundation for Statistical Computing 
  Platform: x86_64-pc-linux-gnu/x86_64 (64-bit) 
   
  $ /usr/local/pkg/R-3.2-branch-20160718/bin/R --vanilla --no-restore --no-save --silent 
  > splitString <- function(...) { print("Test, do nothing") } 
  > invisible(tools::toTitleCase) 
  Warning message: 
  failed to assign RegisteredNativeSymbol for splitString to splitString since splitString is already defined in the 'tools' namespace  

Another way to trigger that warning is by loading the knitr package, e.g.:

  > require("knitr") 
  Loading required package: knitr 
  Warning: failed to assign RegisteredNativeSymbol for splitString to splitString since splitString is already defined in the 'tools' namespace 

The warning only happens the FIRST time I run any code that triggers it.
To get it to happen again, I need to restart R.

R 3.1.0 and all earlier versions do not throw that warning, because
they do not have any splitString C function (see below) at all.  R
3.2.5 does throw the warning, and I believe 3.3 and all later versions
of R do also (but I cannot currently test that on this machine).

In my case, normally I start R without "--vanilla", and load various
custom libraries of my own, one of which contained an R function
"splitString".  That gave the exact same symptoms as the simpler way
of reproducing the warning above.  In practice, I solved the problem
by renaming my "splitString" function to something else.  But I still
wonder what exactly was going on with that warning.

I noticed that the toTitleCase() R code calls .Call() with a bare
splitString identifier, no quotes around it:

  $ grep -n splitString R-3-[234]*/src/library/tools/R/utils.R
  R-3-2-branch/src/library/tools/R/utils.R:1988:        xx <- .Call(splitString, x, ' -/"()')
  R-3-3-branch/src/library/tools/R/utils.R:2074:        xx <- .Call(splitString, x, ' -/"()\n')
  R-3-4-trunk/src/library/tools/R/utils.R:2074:        xx <- .Call(splitString, x, ' -/"()\n')

  $ find R-3-4-trunk -name .svn -prune -o -type f -print0 | xargs -0 grep -n splitString
  R-3-4-trunk/src/library/tools/R/utils.R:2074:        xx <- .Call(splitString, x, ' -/"()\n')
  R-3-4-trunk/src/library/tools/src/text.c:264:SEXP splitString(SEXP string, SEXP delims)
  R-3-4-trunk/src/library/tools/src/tools.h:45:SEXP splitString(SEXP string, SEXP delims);
  R-3-4-trunk/src/library/tools/src/init.c:53:    CALLDEF(splitString, 2),

Doing that is perfectly legal according to help(".Call"), and
interestingly, it apparently does NOT matter whether that code puts
quotes around the splitString or not - I tried it, and it made no
difference.

Is it generally the case the users MUST NOT define R functions with
the same names as "registered" C functions?  Will something break if
we do?

-- 
Andrew Piskorski <atp at piskorski.com>