require_realative calls self.require instead of Kernel.require

[lasso]

Environment

jruby 9.0.5.0 (2.2.3) 2016-01-26 7bee00d Java HotSpot(TM) 64-Bit Server VM 25.91-b14 on 1.8.0_91-b14 +jit [linux-amd64]
Linux linuxmint 4.4.0-31-generic #50-Ubuntu SMP Wed Jul 13 00:07:12 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux (running in VirtualBox)

Originally discovered by test run on Travis CI: https://travis-ci.org/lasso/racket/jobs/149817357

Expected Behavior

MRI/Rubinius behavior:
When using require_relative in a module Foo that also have a module function named require, require_relative (from Kernel) calls Kernel.require, not Foo.require.

When running main.rb (attached) using MRI, the version string '0.5.0' is printed.

Actual Behavior

JRuby behavior:
When using require_relative in a module Foo that also have a module function named require, require_relative (from Kernel) calls Foo.require, not Kernel.require.

When running main.rb (attached) using JRuby, an error is thrown (because Foo.require is called).

A simple workaround that worked for me was using Kernel.require_relative instead of of require_relative in the Foo.version method, then JRuby uses Kernel.require instead of Foo.require.

In summary MRI (and Rubinius) always calls Kernel.require from Kernel.require_relative, but JRuby seems to use self.require if available.

require-bug.zip

[lasso]

The main difference between JRuby and other implementations seems to be that JRuby is the only one that is defining require_relative on the "ruby" side (thus following ruby rules). Other implementations have custom (internal) classes/functions to require_files.

MRI uses internal C functions: https://github.com/ruby/ruby/blob/c3ceb1bff26a0dc59d7b93647e3a58c57e7c0440/load.c#L833

Rubinius uses an internal class called CodeLoader: https://github.com/rubinius/rubinius/blob/2bf206d50c0fb1916e17d936de2e7a4c2a42145b/core/code_loader.rb#L138

JRuby uses the public Kernel class:

jruby/core/src/main/ruby/jruby/kernel/kernel.rb

Line 3 in 9edd069

def require_relative(relative_arg)

[kares]

@lasso sounds like JRuby might do a "better" Ruby job here :) ... suppose you wanted to override require method in your module in MRI there would need to be special handling (re-def) for require_relative as well. maybe should try escalating the issue to MRI itself (they might accept it as a bug for a next release).

[lasso]

@kares - I agree that JRuby does a better job not "special casing" methods in Kernel, but in my code I overrode require, not require_relative. The docs (http://www.rubydoc.info/stdlib/core/Kernel#require_relative-instance_method) does not say that require_relative is depending on the implementation of require, so I actually think JRuby does the wrong thing here. That dependency is based on implementation and JRuby have chosen a different path here. I'm not saying it's the wrong path, just that it seems to be a less used at the moment.

[headius]

Yeah, we have to fix this. I agree it would be nice to have some uniformity in the redispatch to require, but we can't differ this much :-)