UTF Characters for relative path

[JSMonk]

Version: 10.15.3
Platform: Windows 64bit
Subsystem: "path" module

const path = require("path");

const rootPath = "İ"; // Turkish character, char code 304
const absolute = "c:\\Users\\root\\Documents\\İ\\test.txt";

const relative = path.relative(rootPath, absolute); // est.txt

Expected result: test.txt
Actual result: est.txt

[targos]

The problem here is that path.relative computes everything using the lowercase versions of the paths but returns the result using the original case.
"İ" has one char code while "İ".toLowerCase() has two.

There's an easy fix for this case:

console.log(path.win32.relative('c:\\a\\İ', 'c:\\a\\İ\\test.txt'));
// est.txt

diff --git a/lib/path.js b/lib/path.js
index 2301a6ebb8..e76513f0f3 100644
--- a/lib/path.js
+++ b/lib/path.js
@@ -499,7 +499,7 @@ const win32 = {
         if (to.charCodeAt(toStart + i) === CHAR_BACKWARD_SLASH) {
           // We get here if `from` is the exact base path for `to`.
           // For example: from='C:\\foo\\bar'; to='C:\\foo\\bar\\baz'
-          return toOrig.slice(toStart + i + 1);
+          return toOrig.slice(toStart + i + 1 - from.length + fromOrig.length);
         }
         if (i === 2) {
           // We get here if `from` is the device root.

console.log(path.win32.relative('c:\\a\\İ', 'c:\\a\\İ\\test.txt'));
// test.txt

However this case is trickier:

console.log(path.win32.relative('c:\\İ\\a\\İ', 'c:\\İ\\b\\İ\\test.txt'));
// ..\..b\İ\test.txt

[targos]

/cc @BridgeAR @mscdex @cjihrig

[BridgeAR]

This should be easy to fix: instead of using the lower cased variation it's possible to use the upper case one. These two ways do not work identical and the upper case variation does not show the negative sides as the lower case. I am compiling right now to see if that works as expected.

[BridgeAR]

> console.log(path.win32.relative('c:\\a\\İ', 'c:\\a\\İ\\test.txt'));
test.txt
> console.log(path.win32.relative('c:\\İ\\a\\İ', 'c:\\İ\\b\\İ\\test.txt'));
..\..\b\İ\test.txt

[targos]

@BridgeAR what about: console.log(path.win32.relative('c:\\İ\\a\\İ'.toLowerCase(), 'c:\\İ\\b\\İ\\test.txt'.toLowerCase())); ?

[targos]

Sorry ^. That's not the example I meant. Isn't there the same risk with toUpperCase but with other characters?

[targos]

@BridgeAR Make sure this case works as well: console.log(path.win32.relative('c:\\ß\\a\\ß', 'c:\\ß\\b\\ß\\test.txt'));

[BridgeAR]

@targos ß is converted to SS when using toUpperCase, so it has a similar problem like the initial issue. Unicode is pretty ugly and we have problems with Turkish, German and likely more. There is the toLocaleLowerCase() function and it should work for all cases.

@mathiasbynens PTAL.

[targos]

I'm afraid the behavior of toLocaleLowerCase() depends on the user's locale.

[BridgeAR]

@targos yes, that's correct. As far as I know a device name may also contain unicode characters and if that's the case also path.win32.resolve() would fail (and likely more).

I have to think about this a bit closer when I find to do so. We probably also have issues even without changing the casing since we expect characters to always have a length of 1. This is mostly not an issue, since we almost always just check for slashes and take everything in-between but we expect the device root to always be of length 2.

[sr258]

This is still an issue that can lead to bugs in August 2021. Any news?