Make sure to deal with malformed UTF-8 in file names

[palant]

Description

This streamlines the logic in directory listings as discussed. It also makes sure to operate with file names as OsString, allowing for invalid Unicode in file names. The (potentially lossy) conversion to string only happens when the name needs to be displayed. Quite a few unnecessary error conditions got removed as a side-effect.

Something to consider would be enabling serialization of the uri field in JSON. Since the name field might have lost information, there is currently no reliable way to deduce the file’s URL from the JSON data.

Related Issue

#371

How Has This Been Tested?

On the Linux command line I did:

echo test > `echo -e "file\x80.txt"`
mkdir `echo -e "dir\x80.txt"`
echo test > `echo -e "dir\x80/test.txt"`

I verified that both the directory and the file show up in the listing (with the Unicode placeholder showing up for \x80) and both can be opened. The file text.txt in the directory can be opened as well.

As to Windows, I’m not sure that creating a file with a non-Unicode name is even possible. I did verify that the fallback employed there works on Linux (for properly encoded file names) but I didn’t perform any testing on Windows.

[semanticdiff-com]

Review changes with SemanticDiff.

Analyzed 2 of 2 files.

Overall, the semantic diff is 22% smaller than the GitHub diff.

	Filename	Status
✔️	src/directory_listing.rs	22.1% smaller
✔️	src/file_path.rs	18.64% smaller

[palant]

This seems to have improved the performance of large directory slightly, around 2%.

[palant]

Another observation: with this change symlinks show up with the size of the linked file which is correct IMHO. Previously it was the size of the symlink which makes little sense given that the file is otherwise not indicated as a symlink.

The MIME type is being derived from the file name of the symlink however, not the linked file. I wonder whether that is the right thing to do here.

[joseluisq]

Just minor comments to check.

[joseluisq]

The MIME type is being derived from the file name of the symlink however, not the linked file. I wonder whether that is the right thing to do here.

Good question, yeah the symlink name could differ from the actual linked file but I do not if we should follow the symlink to that point.
Maybe we could come up with some idea of how other web servers handle this.
But, honestly symlink following should have been an opt-in feature and disabled by default.

[joseluisq]

No need to delay the PR.
Thanks!