extra::test::test_lots_of_files appears to fail if there are any non-utf8 file names in the top 4 sublevels of /

[huonw]

It runs glob("/*/*/*/*").skip(10000), which traverses a lot of files and if any of them have non utf8 names, this fails the is_utf8 assertion (presumably when constructing the path).

[alexcrichton]

if str::from_utf8 raised on a condition instead, this would be a pretty good instance to return "Just terminate the string" and then processing could go as normal.

I'm not entirely sure what the status of that is, though.

[thestinger]

It shouldn't silently fail to work if there are non-UTF-8 file paths. Assuming all paths can be stored in strings is incorrect on Rust's part. It's not true on Windows either, because NTFS uses UCS2 and doesn't enforce valid Unicode.

[alexcrichton]

Oh yeah this test specifically would ignore the condition, but the condition handler wouldn't be installed by default by the glob function.

[lilyball]

This should be covered by #7225.

[flaper87]

Based on the result of the proposed PR, this is still an issue.

[flaper87]

Complete test case for this issue

    #[test]
    #[cfg(not(windows))]
    fn test_non_utf8_glob() {
        let dir = tempfile::TempDir::new("").unwrap();
        let p = dir.path().join(&[0xFFu8]);
        fs::mkdir(&p, S_IRWXU as u32);

        let pat = p.with_filename("*");
        assert_eq!(glob(pat.as_str().expect("tmpdir is not utf-8")).collect::<~[Path]>(), ~[p])
    }

[lilyball]

I have filed a new issue specifically about glob() and non-utf8 files, which is the root cause of this one, as #11916. Closing this in favor of that.