Should probably tag @kares

@presidentbeef, have you tried with 9.1.12.0? This worked for me.

9.1.12.0 has the same behavior.

@presidentbeef can you try latest on jruby-9.1.13 branch? @kares added some logic which should work a week or two ago and it did not get associated with this issue.

Ah I knew I had downloaded your test repo but now running it I can see require_relative is broken still although I think @kares patch did fix a good chunk of the issues around the fakepath part of it all. So here is a smaller repro:

relative_arg = "lib/test"
dir = "classpath:/"
p File.expand_path(relative_arg, dir)

This prints out

"classpath:C:/lib/test"

The C: should not be there...

Thanks for checking @enebo.

Ok fixed the letter and now I see this:

"uri:classloader:\\lib\\test.rb"
"classpath:\\jar-bootstrap.rb"

GRRRRRRR, :)

Yay...so FILE is correct but File.realpath is swapping the / to \. So a second weird behavior.

@presidentbeef Is the reason for doing this to potentially elide stuff like '..' or '.'. I don't think classpath uris can actually have softlinks. With that said it no classpath: uri should ever get converted to back slashes. So it should just get fixed.

Our app does have a lot of uses of require_relative "../../ but it would be nice for it to "just work" inside a jar.

@presidentbeef yeah I think you should not need to be concerned with whether you are calling with real filesystem paths or a classpath uri. I just wondered if you were using it for '..'. I am pushing a fix here. I would love it if you could try your product on jruby-9.1 branch and see if it is golden now?

9.1.13.0 works!! Thank you @enebo 😄