[Rd] R for Windows leaves detritus in the temp directory

[Henrik Bengtsson]

ISSUE:

The TMPDIR validation done in src/gnuwin32/system.c:

    /* in case getpid() is not unique -- has been seen under Windows */
    snprintf(ifile, 1024, "%s/Rscript%x%x", tm, getpid(),
         (unsigned int) GetTickCount());
    ifp = fopen(ifile, "w+b");
    if(!ifp) R_Suicide(_("creation of tmpfile failed -- set TMPDIR suitably?"));
  }

does _not_ clean up after itself, i.e. there's a missing

    unlink(ifile);

In contrast, ditto in src/unix/system.c does this.


BACKGROUND:

When running R CMD check --as-cran on my 'future' package, I get:

* checking for detritus in the temp directory ... NOTE
Found the following files/directories:
  'Rscript171866c62e'

when checked on R Under development (unstable) (2021-06-13 r80496),
including on win-builder.  I can reproduce this with a package
'tests/detritus.R':

  cl <- parallel::makeCluster(1)
  dummy <- parallel::clusterEvalQ(cl, {
    cl <- parallel::makeCluster(1)
    on.exit(parallel::stopCluster(cl))
    parallel::clusterEvalQ(cl, Sys.getpid())
  })
  print(dummy)
  parallel::stopCluster(cl)


I believe it requires a nested PSOCK cluster to reproduce the 'R CMD
check' NOTE, e.g. it does _not_ happen with:

  cl <- parallel::makeCluster(1)
  dummy <- parallel::clusterEvalQ(cl, {
    Sys.getpid()
  })
  print(dummy)
  parallel::stopCluster(cl)

/Henrik