SDL_filesystem feature request, SDL_CreateDirectory

[ITotalJustice]

Looking at the current filesystem files, seems that for a few platforms, we already have mkdir-like options

vita:

SDL/src/filesystem/vita/SDL_sysfilesystem.c

Line 84 in c59d4dc

sceIoMkdir(retval, 0777);

emscripten:

SDL/src/filesystem/emscripten/SDL_sysfilesystem.c

Line 80 in c59d4dc

if (mkdir(retval, 0700) != 0 && errno != EEXIST) {

haiku:

SDL/src/filesystem/haiku/SDL_sysfilesystem.cc

Line 101 in c59d4dc

create_directory(retval, 0700); // Haiku api: creates missing dirs

os2:

SDL/src/filesystem/os2/SDL_sysfilesystem.c

Line 108 in c59d4dc

DosCreateDir(acBuf, NULL);

cocoa:

SDL/src/filesystem/cocoa/SDL_sysfilesystem.m

Line 123 in c59d4dc

mkdir(retval, 0700);

windows:

SDL/src/filesystem/windows/SDL_sysfilesystem.c

Line 163 in bc7ac13

api_result = CreateDirectoryW(path, NULL);

riscos:

SDL/src/filesystem/riscos/SDL_sysfilesystem.c

Line 196 in 9ee6942

error = createDirectoryRecursive(dir);

winrt:

SDL/src/filesystem/winrt/SDL_sysfilesystem.cpp

Line 225 in c59d4dc

api_result = CreateDirectoryW(path, NULL);

unix:

SDL/src/filesystem/unix/SDL_sysfilesystem.c

Line 318 in 345c161

if (mkdir(retval, 0700) != 0 && errno != EEXIST) {

here's an example of how the api could look

typedef enum
{
    SDL_DIRECTORY_READ = 1 << 0,
    SDL_DIRECTORY_WRITE = 1 << 1,
    SDL_DIRECTORY_EXECUTE = 1 << 2,

    SDL_DIRECTORY_RW = SDL_DIRECTORY_READ | SDL_DIRECTORY_WRITE,
    SDL_DIRECTORY_RWX = SDL_DIRECTORY_READ | SDL_DIRECTORY_WRITE | SDL_DIRECTORY_EXECUTE,
} SDL_DirectoryFlags;

int SDL_CreateDirectory(const char* path, Uint32 flags);

I'm not sure what the return code should look like. 0 on success probably, anything negative for error, maybe a positive value for warnings such as if the dir already exists?

[ITotalJustice]

Thinking about it, it would probably make sense to also add a SDL_RemoveDirectory(path, flags)? flags would have something like remove if non empty (recursively deleting files / folder, can't think of other options, maybe thats all that's needed

[ITotalJustice]

i added a impl for mkdir platforms, this is based off

SDL/src/filesystem/unix/SDL_sysfilesystem.c

Line 257 in 345c161

SDL_GetPrefPath(const char *org, const char *app)

int SDL_CreateDirectory(const char* path, Uint32 flags) {
    char* str = NULL;
    char *ptr = NULL;
    size_t len = 0;
    int mode = 0;

    if (!path)
        return SDL_InvalidParamError("path");

    if (flags & SDL_DIRECTORY_READ)
        mode |= S_IRUSR | S_IRGRP | S_IROTH;

    if (flags & SDL_DIRECTORY_WRITE)
        mode |= S_IWUSR | S_IWGRP | S_IWOTH;

    if (flags & SDL_DIRECTORY_EXECUTE)
        mode |= S_IXUSR | S_IXGRP | S_IXOTH;

    if (!mode)
        return SDL_InvalidParamError("flags");

    len = SDL_strlen(path);
    if (!len) // sanity check
        return -1;

    str = SDL_strdup(path);
    if (!str)
        return SDL_OutOfMemory();

    if (str[len - 1] == '/')
        str[len - 1] = '\0';

    for (ptr = str+1; *ptr; ptr++) {
        if (*ptr == '/') {
            *ptr = '\0';
            if (mkdir(str, mode) != 0 && errno != EEXIST)
                goto error;
            *ptr = '/';
        }
    }

    if (mkdir(str, mode) != 0 && errno != EEXIST) {
error:
        SDL_SetError("Couldn't create directory '%s': '%s'", str, strerror(errno));
        SDL_free(str);
        return -1;
    }

    SDL_free(str);
    return 0;
}

[AntTheAlchemist]

Can we please also have SDL_RenameFile() and SDL_DeleteFile(). No need for RemoveDirectory() to work if folder isn't empty, that can be left to the user to sort out with SDL_DeleteFile().

[ell1e]

No need for RemoveDirectory() to work if folder isn't empty, that can be left to the user to sort out with SDL_DeleteFile().

That is actually quite difficult to get right, because you can have race conditions particularly if you don't want to descend into symbolic link children that point to other folders: it is very easy to mess it up such that between your symbolic link check on the child and its actual removal right after, there is a possibility for it to have been converted to a symlink pointing to an outside folder just in between, causing you to traverse out and potentially delete tons out of tree you never wanted deleted. (For what it's worth, afaik the only way to avoid it is to use directory/file handles, and check link status & then do removal on the reused opened handle rather than via the path string.) Another problem is that if you go through the multibyte conversion to utf-8 on Windows, I think children with invalid utf-16 names (with broken surrogates) may cause your recursive delete to permanently trip over as well - so the deletion loop/queue in itself should ideally retain/use native wide strings for child access. All in all, I think it'd be very useful to solve this in SDL2 properly because it's hard to get all corner cases right.

[AntTheAlchemist]

It's always been the role of the programmer to clear out a directory before attempting to delete it. That's not something the SDL API should be getting involved with. The ethos is to provide the raw low-level functionality so that the programmer can implement the more complex higher function of clearing the directory out. Symbolic links are a moot point in this.

[PythagoRascal]

This might be outside the scope of this feature request, but I recently wished there was an SDL equivalent to stat. I also at the time would have liked file-path manipulation functions or directory enumeration functions.

[ell1e]

It's always been the role of the programmer to clear out a directory b

I really don't understand your point. If the SDL filesystem doesn't provide raw handles, then it is actually impossible to do this on top of it safely. Symbolic links are absolutely not moot, because symlink escape deletion damage is a very common and possibly significant security hole. (E.g. in all cases where you give a user access to an SDL2 daemon's data folder - think dedicated game server with mods - and the daemon may have more/different permissions than you, and now you can make it delete random things anywhere, where it has write access to of course, via symlinks. Not a good look.) So again, it would be useful to solve this in a safer way for everyone, IMHO. In an ideal world this would be left to all apps themselves, but I just don't think that's a good idea here.

[AntTheAlchemist]

It isn't the role of SDL to traverse directory trees. It will never happen. That's why there aren't any file handling functions already, which is why we're more likely to get the very basic ones, if any.

[ell1e]

I do agree with you in so far that it would probably be better to not ever add a way to delete or enumerate a directory (empty or not) unless the API goes the entire way of adding in a good recursive delete in a thoughtful manner. Otherwise some will inevitably put that on top anyway, and badly.

[baidwwy]

and openfile
https://github.com/crazecoder/open_file

[icculus]

Moving this over to #8129 for concrete API discussion.