File.cpp: revert get_parent_dir change

[elad335]

Should fix #14031 .
Tagging @brian218 to explain the change.
Issues's problem is that std::path("＄locks") returns "＄locks/" with japanese locale.
Keep in mind that we already call setlocal("C") on startup so some code overwrites it.

Some places in code also use the pattern (file_path.substr(get_parent_dir(file2_path_in_x_directory, x_times).size()) == trail path from parent directory) which breaks with the code on master.

[brian218]

The changes introduced in 6a570ae simplified the implementation of the function get_parent_dir() and made it neater while keeping the same aim.
In additional, the original implementation (as well as the one brought back by this PR) seems to be unable to handle or normalize some badly-composed paths.
See the test below:

Test code:

for (int i = 0; i < 7; i++)
{
    std::string test = "/dev_hdd0//./AA//./BB/..///CC///DD/.//EE/FF//////";
    test = get_parent_dir(test, i);
    std::cout << test << std::endl;
}

[brian218]

Also, since the sys_fs stuff now keeps track of the the mount paths passed to sys_fs_mount(), it's really important to properly normalize each paths so that the paths can be correctly compared or looked up in the sys_fs map.

[elad335]

If you want to organize badly origanized paths you may add another function then, it's not this function's purpose.

[brian218]

If you want to orginize badly composed paths you may add another function then, it's not this function's purpose.

But it was even not properly getting the parent dir of some given paths. 😆

[elad335]

Examples or it didn't happen.

[brian218]

Examples or it didn't happen.

As you can see, /BB/.. was never correctly handled in this case.

[elad335]

It's not a bug, both paths mean the same thing, with the difference being filesystems typically also check for BB's existence in case it's symlinking to another path.

[Nekotekina]

I think path normalization should be a separate function