My stab at fixing JRUBY-7122 #674

merged 1 commit into from May 1, 2013


None yet

3 participants

trejkaz commented Apr 29, 2013

This fixes the issue for classpath:/ "relative" paths which by their very nature are always absolute. The same issue could be happening for some other URI schemes so I'm not sure if a special case for "classpath" is the right fix.

headius commented May 1, 2013

Finally dug back down to this. Reviewing.

headius commented May 1, 2013

We could probably expand this to also include jar: and jar-based file: URL forms, so we don't have to patch it in the future. Want to give that a shot?

I think the possible forms are:


and maybe


As I wrote this, it occurred to me that both of those cases will still need realpath logic for since there could be maybe that should be a separate issue/PR.

Anyway, I'll merge in your fix, since it's definitely correct for classpath: entries.

@headius headius merged commit 8964e43 into jruby:master May 1, 2013

1 check passed

default The Travis build passed
trejkaz commented May 7, 2013

I added some specs to my fork for this, but the jar:file: ones fail.

JRubyFile.create is being used to do the relativisation logic. JRubyFile did appear to work for classpath: on this computer (not Windows), but it was giving the wrong result on Windows. Give it a jar:file: path though, and it just explodes because the file doesn't exist.

So either JRubyFile.create would have to be fixed to support non-file-paths or I'd want another utility to do the double-dot removals for me which worked for non-existent paths as well as existing paths.

Then the idea would be to rewrite the method to use the following steps:

  1. Figure out if the relative path is absolute and if it isn't, resolve it relative to the base path. So special cases only when the relative path is absolute, which I assume means starts with a URI prefix or a slash.
  2. Figure out what kind of path we're dealing with:
    • For non-URI paths, use prefix="" and suffix=""
    • If it's (jar:file:).(!.), store off the prefix ($1) and suffix ($2) for later
    • If it's any other URI scheme, return the whole thing as-is immediately (skip canonicalisation)
  3. Canonicalise the bit which does refer to a real file (this method already exists, so it should be painless.)
  4. Slap the prefix and suffix back on and return it.
trejkaz commented May 7, 2013

Something else really odd I'm finding. If I try to build up a path to a jar:file URL where I specify the absolute path, I get an error about the file not existing. If I try to build it up as a relative path from a jar:file URL, that works fine. I don't get why one would work and the other wouldn't.

I improved my fork to pass two more specs and added one more spec - the symlink thing still doesn't work but I'm not entirely sure where the symlink stuff is being done, so it could be an invalid test case.

Also at least one of the tests probably fails on Windows, but I won't know until tomorrow.

trejkaz commented Jun 17, 2013

Finding another user here having issues with require_relative. I'm sure this is probably tied into the same mess. I initially couldn't figure out what was different from my path but then I noticed his path...


I bet it's that %20... yet another edge case I didn't expect.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment