Skip to content
An implementation of C++17 std::filesystem for C++11 /C++14/C++17 on Windows, macOS and Linux.
C++ CMake
Branch: master
Clone or download

Supported Platforms Build Status Build status Coverage Status Latest Release Tag


This is a header-only single-file std::filesystem compatible helper library, based on the C++17 specs, but implemented for C++11, C++14 or C++17 (tightly following the C++17 with very few documented exceptions). It is currently tested on macOS 10.12/10.14, Windows 10, and Ubuntu 18.04 but should work on other versions too, as long as you have a C++11 compatible compiler. It is of course in its own namespace ghc::filesystem to not interfere with a regular std::filesystem should you use it in a mixed C++17 environment.

It could still use some polishing, test coverage is above 90%, I didn't benchmark much yet, but I'll try to optimize some parts and refactor others, so I'm striving to improve it as long as it doesn't introduce additional C++17 compatibility issues. Feedback is always welcome. Simply open an issue if you see something missing or wrong or not behaving as expected and I'll comment.


I'm often in need of filesystem functionality, mostly fs::path, but directory access too, and when beginning to use C++11, I used that language update to try to reduce my third-party dependencies. I could drop most of what I used, but still missed some stuff that I started implementing for the fun of it. Originally I based these helpers on my own coding- and naming conventions. When C++17 was finalized, I wanted to use that interface, but it took a while, to push myself to convert my classes.

The implementation is closely based on chapter 30.10 from the C++17 standard and a draft close to that version is Working Draft N4687. It is from after the standardization of C++17 but it contains the latest filesystem interface changes compared to the Working Draft N4659.

I want to thank the people working on improving C++, I really liked how the language evolved with C++11 and the following standards. Keep on the good work!

Oh, and if you ask yourself, what ghc is standing for, it is simply gulraks helper classes, yeah, I know, not very imaginative, but I wanted a short namespace and I use it in some of my private classes (so it has nothing to do with Haskell).


ghc::filesystem is developed on macOS but tested on Windows and Linux. It should work on any of these with a C++11-capable compiler. I currently don't have a BSD derivate besides macOS, so the preprocessor checks will cry out if you try to use it there, but if there is demand, I can try to help. Also there are some checks to hopefully better work on Android, but as I currently don't test with the Android NDK, I wouldn't call it a supported platform yet. All in all, I don't see it replacing std::filesystem where full C++17 is available, it doesn't try to be a "better" std::filesystem, just a drop-in if you can't use it (with the exception of the UTF-8 preference on Windows).

Unit tests are currently run with:

  • macOS 10.12: Xcode 9.2 (clang-900.0.39.2), GCC 8.1.0, Clang 7.0.0, macOS 10.14: Xcode 10.2
  • Windows: Visual Studio 2017, Visual Studio 2015, Visual Studio 2019, MinGW GCC 6.3 (Win32), GCC 7.2 (Win64)
  • Linux (Ubuntu): GCC (5.5, 6.5, 7.4, 8.1, 8.2), Clang (5.0, 6.0, 7.1, 8.0)


The header comes with a set of unit-tests and uses CMake as a build tool and Catch2 as test framework.

All tests agains this implementation should succeed, depending on your environment it might be that there are some warnings, e.g. if you have no rights to create Symlinks on Windows or at least the test thinks so, but these are just informative.

To build the tests from inside the project directory under macOS or Linux just:

mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Debug ..

This generates filesystem_test, the binary that runs all tests.

If the default compiler is a GCC 8 or newer, or Clang 7 or newer, it additionally tries to build a version of the test binary compiled against GCCs/Clangs std::filesystem implementation, named std_filesystem_test as an additional test of conformance. Ideally all tests should compile and succeed with all filesystem implementations, but in reality, there are some differences in behavior, sometimes due to room for interpretation in in the standard, and there might be issues in these implementations too.



The latest release version is v1.2.4 and source archives can be found here.

Using it as Single-File-Header

As ghc::filesystem is at first a header-only library, it should be enough to copy the header or the include/ghc directory into your project folder oder point your include path to this place and simply include the filesystem.hpp header (or ghc/filesystem.hpp if you use the subdirectory).

Everything is in the namespace ghc::filesystem, so one way to use it only as a fallback could be:

#if defined(__cplusplus) && __cplusplus >= 201703L && defined(__has_include)
#if __has_include(<filesystem>)
#define GHC_USE_STD_FS
#include <filesystem>
namespace fs = std::filesystem;
#ifndef GHC_USE_STD_FS
#include <ghc/filesystem.hpp>
namespace fs = ghc::filesystem;

Note that this code uses a two-stage preprocessor condition because Visual Studio 2015 doesn't like the (<...>) syntax, even if it could cut evaluation early before.

Note also, that on MSVC this detection only works starting from version 15.7 on and when setting the /Zc:__cplusplus compile switch, as the compiler allways reports 199711L without that switch (see).

If you want to also use the fstream wrapper with path support as fallback, you might use:

#if defined(__cplusplus) && __cplusplus >= 201703L && defined(__has_include)
#if __has_include(<filesystem>)
#define GHC_USE_STD_FS
#include <filesystem>
namespace fs {
using namespace std::filesystem;
using ifstream = std::ifstream;
using ofstream = std::ofstream;
using fstream = std::fstream;
#ifndef GHC_USE_STD_FS
#include <ghc/filesystem.hpp>
namespace fs {
using namespace ghc::filesystem;
using ifstream = ghc::filesystem::ifstream;
using ofstream = ghc::filesystem::ofstream;
using fstream = ghc::filesystem::fstream;

Now you have e.g. fs::ofstream out(somePath); and it is either the wrapper or the C++17 std::ofstream.

Be aware, as a header-only library, it is not hiding the fact, that it uses system includes, so they "pollute" your global namespace.

ℹ️ Hint: There is an additional header named ghc/fs_std.hpp that implements this dynamic selection of a filesystem implementation, that you can include instead of ghc/filesystem.hpp when you want std::filesystem where available and ghc::filesystem where not. It also enables the wchar_t support on ghc::filesystem on Windows, so the resulting implementation in the fs namespace will be compatible.

Using it as Forwarding-/Implementation-Header

Alternatively, starting from v1.1.0 ghc::filesystem can also be used by including one of two additional wrapper headers. These allow to include a forwarded version in most places (ghc/fs_fwd.hpp) while hiding the implementation details in a single cpp that includes ghc/fs_impl.hpp to implement the needed code. That way system includes are only visible from inside the cpp, all other places are clean.

Be aware, that it is currently not supported to hide the implementation into a Windows-DLL, as a DLL interface with C++ standard templates in interfaces is a different beast. If someone is willing to give it a try, I might integrate a PR but currently working on that myself is not a priority.

If you use the forwarding/implementation approach, you can still use the dynamic switching like this:

#if defined(__cplusplus) && __cplusplus >= 201703L && defined(__has_include)
#if __has_include(<filesystem>)
#define GHC_USE_STD_FS
#include <filesystem>
namespace fs {
using namespace std::filesystem;
using ifstream = std::ifstream;
using ofstream = std::ofstream;
using fstream = std::fstream;
#ifndef GHC_USE_STD_FS
#include <ghc/fs-fwd.hpp>
namespace fs {
using namespace ghc::filesystem;
using ifstream = ghc::filesystem::ifstream;
using ofstream = ghc::filesystem::ofstream;
using fstream = ghc::filesystem::fstream;

and in the implementation hiding cpp, you might use (before any include that includes ghc/fs_fwd.hpp to take precedence:

#if !(defined(__cplusplus) && __cplusplus >= 201703L && defined(__has_include)
#if __has_include(<filesystem>))
#include <ghc/fs_impl.hpp>

ℹ️ Hint: There are additional helper headers, named ghc/fs_std_fwd.hpp and ghc/fs_std_impl.hpp that use this technique, so you can simply include them if you want to dynamically select the filesystem implementation. they also enable the wchar_t support on ghc::filesystem on Windows, so the resulting implementation in the fs namespace will be compatible.

Git Submodule and CMake

Starting from v1.1.0, it is possible to add ghc::filesystem as a git submodule, add the directory to your CMakeLists.txt with add_subdirectory() and then simply use target_link_libraries(your-target ghc_filesystem) to ensure correct include path that allow #include <ghc/filesystem.hpp> to work.


There is a version macro GHC_FILESYSTEM_VERSION defined in case future changes might make it needed to react on the version, but I don't plan to break anything. It's the version as decimal number (major * 10000 + minor * 100 + patch).

Note: Starting from v1.0.2 only even patch versions will be used for releases and odd patch version will only be used for in between commits while working on the next version.


There is almost no documentation in this release, as any std::filesystem documentation would work, besides the few differences explained in the next section. So you might head over to for a description of the components of this library.

The only additions to the standard are documented here:

ghc::filesystem::ifstream, ghc::filesystem::ofstream, ghc::filesystem::fstream

These are simple wrappers around std::ifstream, std::ofstream and std::fstream. They simply add an open() method and a constuctor with an ghc::filesystem::path argument as the fstream variants in C++17 have them.


This is a helper class that currently checks for UTF-8 encoding on non-Windows platforms but on Windows it fetches the command line arguments als Unicode strings from the OS with

::CommandLineToArgvW(::GetCommandLineW(), &argc)

and then converts them to UTF-8, and replaces argc and argv. It is a guard-like class that reverts its changes when going out of scope.

So basic usage is:

namespace fs = ghc::filesystem;

int main(int argc, char* argv[])
    fs::u8arguments u8guard(argc, argv);
    if(!u8guard.valid()) {
        std::cerr << "Bad encoding, needs UTF-8." << std::endl;

    // now use argc/argv as usual, they have utf-8 enconding on windows
    // ...

    return 0;

That way argv is UTF-8 encoded as long as the scope from main is valid.

Note: On macOS, while debugging under Xcode the code currently will return false as Xcode starts the application with US-ASCII as encoding, no matter what encoding is actually used and even setting LC_ALL in the product scheme doesn't change anything. I still need to investigate this.


As this implementation is based on existing code from my private helper classes, it derived some constraints of it, leading to some differences between this and the standard C++17 API.

LWG Defects

This implementation has switchable behavior for the LWG defects #2682, #2935 and #2937. The currently selected behavior is following #2682, #2937 but not following #2935, as I feel it is a bug to report no error on a create_directory() or create_directories() where a regular file of the same name prohibits the creation of a directory and forces the user of those functions to double-check via fs::is_directory if it really worked. The more intuitive approach to directory creation of treating a file with that name as an error is also advocated by the newer paper WG21 P1164R0, the revison P1161R1 was agreed upon on Kona 2019 meeting see merge and GCC by now switched to following its proposal (GCC #86910).

Not Implemented on C++ before C++17

// methods in ghc::filesystem::path:
path& operator+=(basic_string_view<value_type> x);
int compare(basic_string_view<value_type> s) const;

These are not implemented under C++11 and C++14, as there is no std::basic_string_view available and I did want to keep this implementation self-contained and not write a full C++17-upgrade for C++11/14. Starting with v1.1.0 these are supported when compiling ghc::filesystem under C++17.

Differences in API


In Windows, an implementation should use std::wstring and wchar_t as types used for the native representation, but as I'm a big fan of the "UTF-8 Everywhere" philosophy, I decided agains it for now. If you need to call some Windows API, use the W-variant with the path::wstring() member (e.g. GetFileAttributesW(p.wstring().c_str())). This gives you the Unicode variant independant of the UNICODE macro and makes sharing code between Windows, Linux and macOS easier.

Starting with v1.2.0 ghc::filesystem has the option to select the more standard conforming APi with wchar_t and std::wstring on Windows by defining GHC_WIN_WSTRING_STRING_TYPE. This define has no effect on other platforms and will be set by the helping headers ghc/fs_std.hpp and the pair ghc/fs_std_fwd.hpp/ghc/fs_std_impl.hpp to enhance compatibility.

const path::string_type& path::native() const /*noexcept*/;
const path::value_type *path::c_str() const /*noexcept*/;

These two can not be noexcept with the current implementation. This due to the fact, that internally path is working on the generic path version only, and the getters need to do a conversion to native path format.

const path::string_type& path::generic_string() const;

This returns a const reference, instead of a value, because it can. This implementation uses the generic representation for internal workings, so it's "free" to return that.

Differences in Behavior

fs.path (ref)

As the complete inner mechanics of this implementation fs::path are working on the generic format, it is the internal representation. So creating any mixed slash fs::path object under Windows (e.g. with "C:\foo/bar") will lead to a unified path with "C:\foo\bar" via native() and "C:/foo/bar" via generic_string() API.

Additionally this implementation follows the standards suggestion to handle posix paths of the form "//host/path" and USC path on windows also as having a root-name (e.g. "//host"). The GCC implementation didn't choose to do that while testing on Ubuntu 18.04 and macOS with GCC 8.1.0 or Clang 7.0.0. This difference will show as warnings under std::filesystem. This leads to a change in the algorithm described in the standard for operator/=(path& p) where any path p with p.is_absolute() will degrade to an assignment, while this implementation has the exception where *this == *this.root_name() and p == preferred_seperator a normal append will be done, to allow:

fs::path p1 = "//host/foo/bar/file.txt";
fs::path p2;
for (auto p : p1) p2 /= p;
ASSERT(p1 == p2);

For all non-host-leading paths the behaviour will match the one described by the standard.

fs.op.copy (ref)

Then there is fs::copy. The tests in the suite fail partially with C++17 std::filesystem on GCC/Clang. They complain about a copy call with fs::copy_options::recursive combined with fs::copy_options::create_symlinks or fs::copy_options::create_hard_links if the source is a directory. There is nothing in the standard that forbids this combination and it is the only way to deep-copy a tree while only create links for the files. There is LWG #2682 that supports this interpretation, but the issue ignores the usefulness of the combination with recursive and part of the justification for the proposed solution is "we did it so for almost two years". But this makes fs::copy with fs::copy_options::create_symlinks or fs::copy_options::create_hard_links just a more complicated syntax for the fs::create_symlink or fs::create_hardlink operation and I don't want to believe, that this was the intention of the original writing. As there is another issue related to copy, with a different take on the description.

Note: With v1.1.2 I decided to integrate a behavior switch for this and make the LWG #2682 the default.

Open Issues

General Known Issues

There are still some methods that break the noexcept clause, some are related to LWG defects, some are due to my implementation. I work on fixing the later ones, and might in cases where there is no way of implementing the feature without risk of an exception, break conformance and remove the noexcept.


Symbolic Links on Windows

As symbolic links on Windows, while being supported more or less since Windows Vista (with some strict security constraints) and fully since some earlier build of Windows 10, when "Developer Mode" is activated, are at time of writing (2018) rarely used, still they are supported with this implementation.


The Windows ACL permission feature translates badly to the POSIX permission bit mask used in the interface of C++17 filesystem. The permissions returned in the file_status are therefore currently synthesized for the user-level and copied to the group- and other-level. There is still some potential for more interaction with the Windows permission system, but currently setting or reading permissions with this implementation will most certainly not lead to the expected behavior.

Release Notes

v1.2.5 (wip)

  • Pull request #23, tests and examples can now be disabled in CMake via seting BUILD_TESTING and BUILD_EXAMPLES to NO, OFF or FALSE.
  • Pull request #25, missing specialization for construction from std::string_view when available was added.
  • Additional test case when std::string_view is available.


  • Enabled stronger warning switches and resulting fixed issues on GCC and MinGW
  • Bugfix for #22, the fs::copy_options where not forwarded from fs::copy to fs::copy_file in one of the cases.


  • Fix for (#21), when compiling on Alpine Linux with musl instead of glibc, the wrong strerror_r signature was expected. The complex preprocessor define mix was dropped in favor of the usual dispatch by overloading a unifying wrapper.


  • Added MinGW 32/64 and Visual Studio 2015 builds to the CI configuration.
  • Fixed additional compilation issues on MinGW.
  • Pull request (#13), set minimum required CMake version to 3.7.2 (as in Debian 8).
  • Pull request (#14), added support for a make install target.
  • Bugfix for (#15), the forward/impl way of using ghc::filesystem missed a <vector> include in the windows case.
  • Bugfix for (#16), VS2019 didn't like the old size dispatching in the utf8 decoder, so it was changed to a sfinae based approach.
  • New feature (#17), optional support for standard conforming wchar_t/std::wstring interface when compiling on Windows with defined GHC_WIN_WSTRING_STRING_TYPE, this is default when using the ghc/fs_std*.hpp header, to enhance compatibility.
  • New feature (#18), optional filesystem exceptions/errors on unicode errors with defined GHC_RAISE_UNICODE_ERRORS (instead of replacing invalid code points or UTF-8 encoding errors with the replacement character U+FFFD).
  • Pull request (#20), fix for file handle leak in fs::copy_file.
  • Coverage now checked in CI (~95% line coverage).


  • Additional Bugfix for (#12), error in old unified readdir/readdir_r code of fs::directory_iterator; as readdir_r is now depricated, I decided to drop it and the resulting code is much easier, shorter and due to more refactoring faster
  • Fix for crashing unit tests against MSVC C++17 std::filesystem
  • Travis-CI now additionally test with Xcode 10.2 on macOS
  • Some minor refactorings


  • Bugfix for (#11), fs::path::lexically_normal() had some issues with ".."-sequences.
  • Bugfix for (#12), fs::recursive_directory_iterator could run into endless loops, the methods depth() and pop() had issues and the copy behaviour and input_iterator_tag conformance was broken, added tests
  • Restructured some CMake code into a macro to ease the support for C++17 std::filesystem builds of tests and examples for interoperability checks.
  • Some fixes on Windows tests to ease interoperability test runs.
  • Reduced noise on fs::weakly_canonical() tests against std::fs
  • Added simple du example showing the recursive_directory_iterator used to add the sizes of files in a directory tree.
  • Added error checking in fs::file_time_type test helpers
  • fs::copy() now conforms LWG #2682, disallowing the use of `copy_option::create_symlinks' to be used on directories


  • Restructuring of the project directory. The header files are now using hpp as extension to be marked as c++ and they where moved to include/ghc/ to be able to include by <ghc/filesystem.hpp> as the former include name might have been to generic and conflict with other files.
  • Better CMake support: ghc::filesystem now can be used as a submodul and added with add_subdirectory and will export itself as ghc_filesystem target. To use it, only target_link_libraries(your-target ghc_filesystem) is needed and the include directories will be set so #include <ghc/filesystem.hpp> will be a valid directive. Still you can simply only add the header file to you project and include it from there.
  • Enhancement (#10), support for separation of implementation and forwarded api: Two additional simple includes are added, that can be used to forward ghc::filesystem declarations (fs_fwd.hpp) and to wrap the implementation into a single cpp (fs_impl.hpp)
  • The std::basic_string_view variants of the fs::path api are now supported when compiling with C++17.
  • Added CI integration for Travis-CI and Appveyor.
  • Fixed MinGW compilation issues.
  • Added long filename support for Windows.


  • Bugfix for (#9), added missing return statement to ghc::filesystem::path::generic_string()
  • Added checks to hopefully better compile against Android NDK. There where no tests run yet, so feedback is needed to actually call this supported.
  • filesystem.h was renamed filesystem.hpp to better reflect that it is a c++ language header.


  • Bugfix for (#6), where ghc::filesystem::remove() and ghc::filesystem::remove_all() both are now able to remove a single file and both will not raise an error if the path doesn't exist.
  • Merged pull request (#7), a typo leading to setting error code instead of comparing it in ghc::filesystem::remove() under Windows.
  • Bugfix for ((#8), the Windows version of ghc::filesystem::directory_iterator now releases resources when reaching end() like the POSIX one does.


  • Bugfix for (#4), missing error_code propagation in ghc::filesystem::copy() and ghc::filesystem::remove_all fixed.
  • Bugfix for (#5), added missing std namespace in ghc::filesystem::recursive_directory_iterator::difference_type.


  • Bugfix for (#3), fixed missing inlines and added test to ensure including into multiple implementation files works as expected.
  • Building tests with -Wall -Wextra -Werror and fixed resulting issues.


  • Updated catch2 to v2.4.0.
  • Refactored fs.op.permissions test to work with all tested std::filesystem implementations (gcc, clang, msvc++).
  • Added helper class ghc::filesystem::u8arguments as argv converter, to help follow the UTF-8 path on windows. Simply instantiate it with argc and argv and it will fetch the Unicode version of the command line and convert it to UTF-8. The destructor reverts the change.
  • Added examples folder with hopefully some usefull example usage. Examples are tested (and build) with ghc::filesystem and C++17 std::filesystem when available.
  • Starting with this version, only even patch level versions will be tagged and odd patch levels mark in-between non-stable wip states.
  • Tests can now also be run against MS version of std::filesystem for comparison.
  • Added missing fstream include.
  • Removed non-conforming C99 timespec/timeval usage.
  • Fixed some integer type mismatches that could lead to warnings.
  • Fixed chrono conversion issues in test and example on clang 7.0.0.


  • Bugfix: ghc::filesystem::canonical now sees empty path as non-existant and reports an error. Due to this ghc::filesystem::weakly_canonical now returns relative paths for non-existant argument paths. (#1)
  • Bugfix: ghc::filesystem::remove_all now also counts directories removed (#2)
  • Bugfix: recursive_directory_iterator tests didn't respect equality domain issues and dereferencable constraints, leading to fails on std::filesystem tests.
  • Bugfix: Some noexcept tagged methods and functions could indirectly throw exceptions due to UFT-8 decoding issues.
  • std_filesystem_test is now also generated if LLVM/clang 7.0.0 is found.


This was the first public release version. It implements the full range of C++17 std::filesystem, as far as possible without other C++17 dependencies.

You can’t perform that action at this time.