Fastest way to check if a file exists using standard C++/C++11,14,17/C?
C++CFileStreamC++ Problem Overview
I would like to find the fastest way to check if a file exists in standard C++11, 14, 17, or C. I have thousands of files and before doing something on them I need to check if all of them exist. What can I write instead of /* SOMETHING */ in the following function?
inline bool exist(const std::string& name)
{
/* SOMETHING */
}
C++ Solutions
Solution 1 - C++
Well I threw together a test program that ran each of these methods 100,000 times, half on files that existed and half on files that didn't.
#include <sys/stat.h>
#include <unistd.h>
#include <string>
#include <fstream>
inline bool exists_test0 (const std::string& name) {
ifstream f(name.c_str());
return f.good();
}
inline bool exists_test1 (const std::string& name) {
if (FILE *file = fopen(name.c_str(), "r")) {
fclose(file);
return true;
} else {
return false;
}
}
inline bool exists_test2 (const std::string& name) {
return ( access( name.c_str(), F_OK ) != -1 );
}
inline bool exists_test3 (const std::string& name) {
struct stat buffer;
return (stat (name.c_str(), &buffer) == 0);
}
Results for total time to run the 100,000 calls averaged over 5 runs,
| Method | Time |
|---|---|
exists_test0 (ifstream) |
0.485s |
exists_test1 (FILE fopen) |
0.302s |
exists_test2 (posix access()) |
0.202s |
exists_test3 (posix stat()) |
0.134s |
The stat() function provided the best performance on my system (Linux, compiled with g++), with a standard fopen call being your best bet if you for some reason refuse to use POSIX functions.
Solution 2 - C++
Remark : in C++14 and as soon as the filesystem TS will be finished and adopted, the solution will be to use:
std::experimental::filesystem::exists("helloworld.txt");
and since C++17, only:
std::filesystem::exists("helloworld.txt");
Solution 3 - C++
I use this piece of code, it works OK with me so far. This does not use many fancy features of C++:
bool is_file_exist(const char *fileName)
{
std::ifstream infile(fileName);
return infile.good();
}
Solution 4 - C++
For those who like boost:
boost::filesystem::exists(fileName)
or, since ISO C++17:
std::filesystem::exists(fileName)
Solution 5 - C++
It depends on where the files reside. For instance, if they are all supposed to be in the same directory, you can read all the directory entries into a hash table and then check all the names against the hash table. This might be faster on some systems than checking each file individually. The fastest way to check each file individually depends on your system ... if you're writing ANSI C, the fastest way is fopen because it's the only way (a file might exist but not be openable, but you probably really want openable if you need to "do something on it"). C++, POSIX, Windows all offer additional options.
While I'm at it, let me point out some problems with your question. You say that you want the fastest way, and that you have thousands of files, but then you ask for the code for a function to test a single file (and that function is only valid in C++, not C). This contradicts your requirements by making an assumption about the solution ... a case of the XY problem. You also say "in standard c++11(or)c++(or)c" ... which are all different, and this also is inconsistent with your requirement for speed ... the fastest solution would involve tailoring the code to the target system. The inconsistency in the question is highlighted by the fact that you accepted an answer that gives solutions that are system-dependent and are not standard C or C++.
Solution 6 - C++
Without using other libraries, I like to use the following code snippet:
#ifdef _WIN32
#include <io.h>
#define access _access_s
#else
#include <unistd.h>
#endif
bool FileExists( const std::string &Filename )
{
return access( Filename.c_str(), 0 ) == 0;
}
This works cross-platform for Windows and POSIX-compliant systems.
Solution 7 - C++
Same as suggested by PherricOxide but in C
#include <sys/stat.h>
int exist(const char *name)
{
struct stat buffer;
return (stat (name, &buffer) == 0);
}
Solution 8 - C++
inline bool exist(const std::string& name)
{
ifstream file(name);
if(!file) // If the file was not found, then file is 0, i.e. !file=1 or true.
return false; // The file was not found.
else // If the file was found, then file is non-0.
return true; // The file was found.
}
Solution 9 - C++
I need a fast function that can check if a file is exist or not and PherricOxide's answer is almost what I need except it does not compare the performance of boost::filesystem::exists and open functions. From the benchmark results we can easily see that :
-
Using stat function is the fastest way to check if a file is exist. Note that my results are consistent with that of PherricOxide's answer.
-
The performance of boost::filesystem::exists function is very close to that of stat function and it is also portable. I would recommend this solution if boost libraries is accessible from your code.
Benchmark results obtained with Linux kernel 4.17.0 and gcc-7.3:
2018-05-05 00:35:35
Running ./filesystem
Run on (8 X 2661 MHz CPU s)
CPU Caches:
L1 Data 32K (x4)
L1 Instruction 32K (x4)
L2 Unified 256K (x4)
L3 Unified 8192K (x1)
--------------------------------------------------
Benchmark Time CPU Iterations
--------------------------------------------------
use_stat 815 ns 813 ns 861291
use_open 2007 ns 1919 ns 346273
use_access 1186 ns 1006 ns 683024
use_boost 831 ns 830 ns 831233
Below is my benchmark code:
#include <string.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <dirent.h>
#include <fcntl.h>
#include <unistd.h>
#include "boost/filesystem.hpp"
#include <benchmark/benchmark.h>
const std::string fname("filesystem.cpp");
struct stat buf;
// Use stat function
void use_stat(benchmark::State &state) {
for (auto _ : state) {
benchmark::DoNotOptimize(stat(fname.data(), &buf));
}
}
BENCHMARK(use_stat);
// Use open function
void use_open(benchmark::State &state) {
for (auto _ : state) {
int fd = open(fname.data(), O_RDONLY);
if (fd > -1) close(fd);
}
}
BENCHMARK(use_open);
// Use access function
void use_access(benchmark::State &state) {
for (auto _ : state) {
benchmark::DoNotOptimize(access(fname.data(), R_OK));
}
}
BENCHMARK(use_access);
// Use boost
void use_boost(benchmark::State &state) {
for (auto _ : state) {
boost::filesystem::path p(fname);
benchmark::DoNotOptimize(boost::filesystem::exists(p));
}
}
BENCHMARK(use_boost);
BENCHMARK_MAIN();
Solution 10 - C++
Another 3 options under windows: #1
inline bool exist(const std::string& name)
{
OFSTRUCT of_struct;
return OpenFile(name.c_str(), &of_struct, OF_EXIST) != INVALID_HANDLE_VALUE && of_struct.nErrCode == 0;
}
#2
inline bool exist(const std::string& name)
{
HANDLE hFile = CreateFile(name.c_str(), GENERIC_READ, 0, NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL);
if (hFile != NULL && hFile != INVALID_HANDLE)
{
CloseFile(hFile);
return true;
}
return false;
}
#3
inline bool exist(const std::string& name)
{
return GetFileAttributes(name.c_str()) != INVALID_FILE_ATTRIBUTES;
}
Solution 11 - C++
You may also do bool b = std::ifstream('filename').good();. Without the branch instructions(like if) it must perform faster as it needs to be called thousands of times.
Solution 12 - C++
If you need to distinguish between a file and a directory, consider the following which both use stat which the fastest standard tool as demonstrated by PherricOxide:
#include <sys/stat.h>
int FileExists(char *path)
{
struct stat fileStat;
if ( stat(path, &fileStat) )
{
return 0;
}
if ( !S_ISREG(fileStat.st_mode) )
{
return 0;
}
return 1;
}
int DirExists(char *path)
{
struct stat fileStat;
if ( stat(path, &fileStat) )
{
return 0;
}
if ( !S_ISDIR(fileStat.st_mode) )
{
return 0;
}
return 1;
}
Solution 13 - C++
You can use std::ifstream, funcion like is_open, fail, for example as below code (the cout "open" means file exist or not):
cited from this answer
Solution 14 - C++
All of the other answers focus on individually checking every file, but if the files are all in one directory (folder), it might be much more efficient to just read the directory and check for the existence of every file name you want.
This might even be more efficient even if the files are spread across several directories, depends on the exact ratio of directories to files. Once you start getting closer to each target file being in its own directory, or there being lots of other files in the same directories which you don't want to check for, then I'd expect it to finally tip over into being less efficient than checking each file individually.
A good heuristic: working on a bunch of data you already have is much faster than asking the operating system for any amount of data. System call overhead is huge relative to individual machine instructions. So it is almost always going to be faster to ask the OS "give me the entire list of files in this directory" and then to dig through that list, and slower to ask the OS "give me information on this file", "okay now give me information on this other file", "now give me information on ...", and so on.
Every good C library implements its "iterate over all files in a directory" APIs in an efficient way, just like buffered I/O - internally it reads up a big list of directory entries from the OS at once, even though the APIs look like asking the OS for each entry individually.
So if I had this requirement, I would
- do everything possible to encourage the design and usage so that all the files were in one folder, and no other files were in that folder,
- put the list of file names that I need to be present into a data structure in memory that has O(1) or at least O(log(n)) lookup and delete times (like a hash map or a binary tree),
- list the files in that directory, and "check off" (delete) each one as I went from the "list" (hash map or binary tree) in memory.
Except depending on the exact use case, maybe instead of deleting entries from a hash map or tree, I would keep track of a "do I have this file?" boolean for each entry, and figure out a data structure that would make it O(1) to ask "do I have every file?". Maybe a binary tree but the struct for each non-leaf node also has a boolean that is a logical-and of the booleans of its leaf nodes. That scales well - after setting a boolean in a leaf node, you just walk up the tree and set each node's "have this?" boolean with the && of its child node's boolean (and you don't need to recurse down those other child nodes, since if you're doing this process consistently every time you go to set one of the leaves to true, they will be set to true if and only if all of their children are.)
Sadly, there is no standard way to do it until C++17.
C++17 got std::filesystem::directory_iterator.
Of course there is a corresponding boost::filesystem::directory_iterator which I presume will work in older versions of C++.
The closest thing to a standard C way is opendir and readdir from dirent.h. That is a standard C interface, it's just standardized in POSIX and not in the C standard itself. It comes is available out-of-the-box on Mac OS, Linux, all the BSDs, other UNIX/UNIX-like systems, and any other POSIX/SUS system. For Windows, there is a dirent.h implementation that you just have to download and drop into your include path.
However, since you're looking for the fastest way, you might want to look beyond the portable/standard stuff.
On Linux, you might be able to optimize your performance by manually specifying the buffer size with the raw system call getdents64.
On Windows, after a bit of digging, it looks like for maximum performance you want to use FindFirstFileEx with FindExInfoBasic and FIND_FIRST_EX_LARGE_FETCH when you can, which a lot of the open source libraries like the above dirent.h for Windows don't seem to do. But for code that needs to work with stuff older than the last couple Windows versions, you might as well just use the straightforward FindFirstFile without the extra flags.
Plan 9 won't be covered by any of the above, and there you'll want dirread or dirreadall (the latter if you can safely assume you have enough memory for the entire directory contents). If you want more control over the buffer size for performance use plain read or read and decode the directory entry data - they're in a documented machine-independent format and I think there's helper functions provided.
I don't know about any other operating systems.
I might edit this answer with some tests later. Others are welcome to edit in test results as well.
Solution 15 - C++
all_of (begin(R), end(R), [](auto&p){ exists(p); })
where R is your sequence of path-like things, and exists() is from the future std or current boost. If you roll your own, keep it simple,
bool exists (string const& p) { return ifstream{p}; }
The branched solution isn't absolutely terrible and it won't gobble file descriptors,
bool exists (const char* p) {
#if defined(_WIN32) || defined(_WIN64)
return p && 0 != PathFileExists (p);
#else
struct stat sb;
return p && 0 == stat (p, &sb);
#endif
}
Solution 16 - C++
In C++17 :
#include <experimental/filesystem>
bool is_file_exist(std::string& str) {
namespace fs = std::experimental::filesystem;
fs::path p(str);
return fs::exists(p);
}
Solution 17 - C++
well there's an even easier way
#include <fstream>
#include <iostream>
void FileExists(std::string myfile){
std::ifstream file(myfile.c_str());
if (file) {
std::cout << "file exists" << std::endl;
}
else {
std::cout << "file doesn't exist" << std::endl;
}
}
int main() {
FileExists("myfile.txt");
return 0;
}
Solution 18 - C++
Here is a simple example!
#include <iostream>
#include <fstream>
using namespace std;
void main(){
SearchFile("test.txt");
}
bool SearchFile(const char *file)
{
ifstream infile(file);
if (!infile.good())
{
// If file is not there
exit(1);
}
}
Solution 19 - C++
The fastest and the safest way to test for file existence is not to separately/explicitly test for it at all. That is, see if you can find a way to replace the ordinary
if(exists(file)) { /* point A */
/* handle existence condition */
return;
}
do_something_with(file); /* point B */
with the improved
r = do_something_with_unless_exists(file);
if(r == 0)
success;
else if(errno == EEXIST)
/* handle existence condition */
else
/* handle other error */
Besides being faster, this eliminates the race condition (specifically, "TOC/TOU") inherent in the first solution, namely the possibility that the file comes into existence between point A and point B.
Obviously the second solution presupposes the existence of an atomic way to do the do_something_with_unless_exists operation. Often there is a way, but sometimes you have to hunt around for it.
-
Creating a file: call
open()withO_CREATandO_EXCL. -
Creating a file in pure C, if you have C11: call
fopen()with"wx". (I only just learned about this one yesterday.) -
Making a directory: just call
mkdir()and check forerrno == EEXISTafterwards. -
Acquiring a lock: any locking system worth its salt already has an atomic acquire-the-lock-as-long-as-nobody-else-has-it primitive.
(There are others, but those are the ones I can think of just now.)
[Footnote: In the early days of Unix, there were no specific, dedicated facilities made available to ordinary processes to do locking, so if you wanted to set up a mutex, this was typically implemented by creating a certain empty directory, since the mkdir syscall has always had the ability to fail or succeed based on prior existence or nonexistence, atomically.]
Solution 20 - C++
Using MFC it is possible with the following
CFileStatus FileStatus;
BOOL bFileExists = CFile::GetStatus(FileName,FileStatus);
Where FileName is a string representing the file you are checking for existance
Solution 21 - C++
there is only one faster way to check if the file exists and if you have permission to read it the way is using C language wish is faster and can be used also in any version in C++
solution: in C there is a library errno.h which has an external (global) integer variable called errno which contains a number that can be used to recognize the type of error
#include <stdio.h>
#include <stdbool.h>
#include <errno.h>
bool isFileExist(char fileName[]) {
FILE *fp = fopen(fileName, "r");
if (fp) {
fclose(fp);
return true;
}
return errno != ENOENT;
}
bool isFileCanBeRead(char fileName[]) {
FILE *fp = fopen(fileName, "r");
if (fp) {
fclose(fp);
return true;
}
return errno != ENOENT && errno != EPERM;
}
Solution 22 - C++
Detecting if a file exists in windows.
bool DoesExistFile(std::wstring filePath)
{
bool result = true;
HANDLE fileHandle = CreateFile(
filePath.c_str(),
GENERIC_READ,
0,
NULL,
OPEN_EXISTING,
0,
NULL);
if ((fileHandle != NULL) && (fileHandle != INVALID_HANDLE_VALUE))
CloseHandle(fileHandle);
else
{
DWORD error = GetLastError();
if ((error == ERROR_FILE_NOT_FOUND) || (error == ERROR_PATH_NOT_FOUND))
result = false;
}
return result;
}
Solution 23 - C++
Although there are several ways to do this the most efficient solution to your problem would probably be to use one of the fstream's predefined method such as good(). With this method you can check whether the file you've specified exist or not.
fstream file("file_name.txt");
if (file.good())
{
std::cout << "file is good." << endl;
}
else
{
std::cout << "file isnt good" << endl;
}
I hope you find this useful.