Having a test for it, would make me more comfortable.

Same for me, actually… I'll try to add one the next time I'll boot my Windows machine.

Ok, I added a test… and now I'm shared about merging this 😬

File.getCanonicalFile depends on what's on disk, so that two calls of os.Path(…).toString may not return the same value (if we pass it a short form, it will return a longer one, but only when the corresponding file exists), which could be surprising.

But a worse issue is that it seems that File.getCanonicalFile caches its values: if it's called with a short form before the file exists, it will later always return the short form, even if we create the underlying file later on.

Path.normalize doesn't suffer those quirks, AFAIK.

And fwiw, a similar issue can be reproduced on macOS (with its "case-insensitive but case-preserving" file systems), using just the case:

$ mkdir Foo && amm -c 'println((os.pwd / "foo").toIO.getCanonicalFile)'
/Users/alexandre/test/Foo

$ rmdir Foo && amm -c 'println((os.pwd / "foo").toIO.getCanonicalFile); os.makeDir(os.pwd / "Foo"); println((os.pwd / "foo").toIO.getCanonicalFile)'
/Users/alexandre/test/foo
/Users/alexandre/test/foo

$ ls
Foo

Ok, I added a test… and now I'm shared about merging this grimacing

Do you mean "scared"?

File.getCanonicalFile depends on what's on disk, so that two calls of os.Path(…).toString may not return the same value (if we pass it a short form, it will return a longer one, but only when the corresponding file exists), which could be surprising.

Oh, as I commented on #78, I'd rather remove any dependency on disk state, and I was under the impression, you're going to always normalize to the long form. In that regard, I don't understand the test case, at least, it test something different.

But a worse issue is that it seems that File.getCanonicalFile caches its values: if it's called with a short form before the file exists, it will later always return the short form, even if we create the underlying file later on.

Path.normalize doesn't suffer those quirks, AFAIK.

Better, we don't merge this.

It'd be better not to merge, yes.

(And it seems there's a need for a library providing true canonical paths, without those caching quirks…)