Faster alternative to file.exists()

Question

I maintain an R package that needs to check the existence of lots of little files individually. Repeated calls to file.exists() produce noticeable slowness ( benchmarking results here ). Unfortunately, situational constraints prevent me from calling file.exists() once on the entire batch of files in vectorized fashion, which I believe would be a lot faster. Is there a faster way to check for the existence of a single file? Maybe in C? This way does not seem to be any faster on my system (the same one that produced these benchmarks ):

library(inline)
library(microbenchmark)

body <- "
  FILE *fp = fopen(CHAR(STRING_ELT(r_path, 0)), \"r\");
  SEXP result = PROTECT(allocVector(INTSXP, 1));
  INTEGER(result)[0] = fp == NULL? 0 : 1;
  UNPROTECT(1);
  return result;
"

file_exists_c <- cfunction(sig = signature(r_path = "character"), body = body)

tmp <- tempfile()

microbenchmark(
  c = file_exists_c(tmp),
  r = file.exists(tmp)
)
#> Unit: microseconds
#>  expr   min     lq    mean median     uq    max neval
#>     c 4.912 5.0230 5.42443 5.0605 5.1240 25.264   100
#>     r 3.972 4.0525 4.32615 4.1835 4.2675 11.750   100

file.create(tmp)
#> [1] TRUE

microbenchmark(
  c = file_exists_c(tmp),
  r = file.exists(tmp)
)
#> Unit: microseconds
#>  expr    min      lq     mean  median      uq    max neval
#>     c 16.212 16.6245 17.04727 16.7645 16.9860 32.207   100
#>     r  6.242  6.4175  7.16057  7.2830  7.4605 26.781   100

^{Created on 2019-12-06 by the reprex package (v0.3.0)}

Edit: access()

access() does appear to be faster, but not by very much.

library(inline)
library(microbenchmark)

body <- "
  SEXP result = PROTECT(allocVector(INTSXP, 1));
  INTEGER(result)[0] = access(CHAR(STRING_ELT(r_path, 0)), 0)? 0 : 1;
  UNPROTECT(1);
  return result;
"

file_exists_c <- cfunction(
  sig = signature(r_path = "character"),
  body = body,
  includes = "#include <unistd.h>"
)

tmp <- tempfile()

microbenchmark(
  c = file_exists_c(tmp),
  r = file.exists(tmp)
)
#> Unit: microseconds
#>  expr   min    lq    mean median     uq    max neval
#>     c 1.033 1.048 1.21334 1.0745 1.0910 13.793   100
#>     r 1.051 1.068 1.19280 1.0930 1.1175 10.048   100

file.create(tmp)
#> [1] TRUE

microbenchmark(
  c = file_exists_c(tmp),
  r = file.exists(tmp)
)
#> Unit: microseconds
#>  expr   min     lq    mean median     uq    max neval
#>     c 1.073 1.0910 1.33543 1.1285 1.1500 16.676   100
#>     r 1.172 1.1965 1.32934 1.2335 1.2695  9.916   100

^{Created on 2019-12-07 by the reprex package (v0.3.0)}

Answer 1

Here is the entirety of file.exists source code (as of this writing):

https://github.com/wch/r-source/blob/bfe73ecd848198cb9b68427cec7e70c40f96bd72/src/main/platform.c#L1375-L1404

SEXP attribute_hidden do_fileexists(SEXP call, SEXP op, SEXP args, SEXP rho)
{
    SEXP file, ans;
    int i, nfile;
    checkArity(op, args);
    if (!isString(file = CAR(args)))
    error(_("invalid '%s' argument"), "file");
    nfile = LENGTH(file);
    ans = PROTECT(allocVector(LGLSXP, nfile));
    for (i = 0; i < nfile; i++) {
    LOGICAL(ans)[i] = 0;
    if (STRING_ELT(file, i) != NA_STRING) {
#ifdef Win32
        /* Package XML sends arbitrarily long strings to file.exists! */
        size_t len = strlen(CHAR(STRING_ELT(file, i)));
        if (len > MAX_PATH)
        LOGICAL(ans)[i] = FALSE;
        else
        LOGICAL(ans)[i] =
            R_WFileExists(filenameToWchar(STRING_ELT(file, i), TRUE));
#else
        // returns NULL if not translatable
        const char *p = translateCharFP2(STRING_ELT(file, i));
        LOGICAL(ans)[i] = p && R_FileExists(p);
#endif
    } else LOGICAL(ans)[i] = FALSE;
    }
    UNPROTECT(1); /* ans */
    return ans;
}

As for R_FileExists , it's here:

https://github.com/wch/r-source/blob/bfe73ecd848198cb9b68427cec7e70c40f96bd72/src/main/sysutils.c#L60-L79

#ifdef Win32
Rboolean R_FileExists(const char *path)
{
    struct _stati64 sb;
    return _stati64(R_ExpandFileName(path), &sb) == 0;
}
#else
Rboolean R_FileExists(const char *path)
{
    struct stat sb;
    return stat(R_ExpandFileName(path), &sb) == 0;
}

( R_ExpandFileName is just doing path.expand ). It's relying on the stat system utility:

https://en.wikipedia.org/wiki/Stat_(system_call)

https://pubs.opengroup.org/onlinepubs/007908799/xsh/sysstat.h.html

It's built for vectorized inputs, so as mentioned it's much preferable to do file.exists(vector_of_files) than to repeatedly run file.exists(single_file) .

From what I can tell (admittedly I'm no expert on the system utilities here), any efficiency gains come at the cost of robustness.

Answer 2

A simple solution in C would be to use access( name of file , 0); if the function returns 0 then the file exists. The second parameter 0 specifies check only if it exists. Example: I check for the file test.txt in /test directory

#include "io.h"
#include "stdio.h"

int main()
{
 if(!access("/test/test.txt",0)) printf("file exists");
}