`FileUtil::relativize` doesn't take symlinks into account

[Krzmbrzl]

JabRef version

5.15 (latest release)

Operating system

GNU / Linux

Details on version and operating system

KDE Neon - based on Ubuntu 24.04

Checked with the latest development build (copy version output from About dialog)

• I made a backup of my libraries before testing the latest development version.
• I have tested the latest development version and the problem persists

Steps to reproduce the behaviour

Look at

jabref/src/main/java/org/jabref/logic/util/io/FileUtil.java

Lines 261 to 263 in af362ea

	if (file.startsWith(directory)) {
	return directory.relativize(file);
	}

Appendix

The effect of this is (among other things?) that when manually linking a PDF file to a bib entry, which is located in the main file directory, it can still end up being added by its full path.

This is because the configured main file directory and/or the path to the selected PDF file can contain symbolic links which make the paths appear different (which causes FileUtils::relativize to fail) even though they are not.

It seems that Path::toRealPath from Java's standard library could be used for obtaining a canonical path representation that has symlinks resolved.

[Siedlerchr]

thanks for the report. I guess another approach would be use a toString comparison

[koppor]

I guess another approach would be use a toString comparison

Example:

• symlink: /home/me/lib/external -> /shared/external
• Library path: /home/me/lib
• PDF name: /shared/external/demo.pdf`
• Intended relative path: external/demo.pdf

How would a string match help here? /home/me/lib/external is not the start of /shared/external. And we should not do any magic match.

[koppor]

@Krzmbrzl Isn't the issue much more complicated?

Crawl down path. Check if its a symlink. If yes: toRealPath. Compare with startsWith of realPath of file?

Otherwise, we cannot handle the case described at #12995 (comment).

We also need to recursively do this, because symlinked directories could contain other symlinked directories.

Reconstructing the real path is also an effort as the symlinks should be used, not the real path.

A test setup can be done with https://github.com/roastedroot/zerofs.

However, we need test cases.

[Krzmbrzl]

Right, there seem to be quite a few possibilities for further complications.
However, I suppose that using the real path on both ends can be used to determine whether or not a relative or an absolute path should be stored (by simple prefix matching). Figuring out the ideal relative path to use will then bring the complications you mentioned.

[koppor]

But what if

• /lib/things -> /share/expensive
• /share/expensive -> /veryprivate
• /veryprivate/a.pdf

If I do realpath, I get

• /lib/bib.bib
  And
• /veryprivate/a.pdf

I don't see how absolute real paths help.

[Krzmbrzl]

Ah true - I had a simpler example in mind but in your case absolute real paths indeed won't help 👀

[Muskan244]

/assign-me