Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

1.7.26 regression: NoMethodError: private method `require' called for Kernel:Module in Sequel #4155

Closed
perlun opened this issue Sep 14, 2016 · 9 comments
Closed

1.7.26 regression: NoMethodError: private method `require' called for Kernel:Module in Sequel #4155

perlun opened this issue Sep 14, 2016 · 9 comments
Labels
core JRuby 1.7.x needs tests
Milestone
JRuby 1.7.27

Comments

@perlun
Copy link
Contributor

perlun commented Sep 14, 2016

Something has gone broken in JRuby 1.7.26, causing the error below.

$ bundle exec irb
The signal IOT is in use by the JVM and will not work correctly on this platform
The signal CLD is in use by the JVM and will not work correctly on this platform
The signal EXIT is in use by the JVM and will not work correctly on this platform
jruby-1.7.26 :001 > require 'sequel'
 => true
jruby-1.7.26 :002 > Sequel.extension :core_extensions
NoMethodError: private method `require' called for Kernel:Module
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/sequel-4.38.0/lib/sequel/core.rb:139:in `extension'
    from org/jruby/RubyArray.java:1613:in `each'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/sequel-4.38.0/lib/sequel/core.rb:139:in `extension'
    from (irb):2:in `evaluate'
    from org/jruby/RubyKernel.java:1079:in `eval'
    from org/jruby/RubyKernel.java:1479:in `loop'
    from org/jruby/RubyKernel.java:1242:in `catch'
    from org/jruby/RubyKernel.java:1242:in `catch'
    from /Users/plundberg/.rvm/rubies/jruby-1.7.26/bin/irb:13:in `(root)'
    from org/jruby/RubyKernel.java:1059:in `load'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/cli/exec.rb:1:in `(root)'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/cli/exec.rb:74:in `kernel_load'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/cli/exec.rb:27:in `run'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/cli.rb:332:in `exec'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/vendor/thor/lib/thor/command.rb:27:in `run'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/vendor/thor/lib/thor/invocation.rb:126:in `invoke_command'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/vendor/thor/lib/thor.rb:359:in `dispatch'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/cli.rb:20:in `dispatch'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/vendor/thor/lib/thor/base.rb:440:in `start'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/cli.rb:11:in `start'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/exe/bundle:34:in `(root)'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/lib/bundler/friendly_errors.rb:100:in `with_friendly_errors'
    from org/jruby/RubyKernel.java:1059:in `load'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/gems/bundler-1.13.1/exe/bundle:26:in `(root)'
    from org/jruby/RubyKernel.java:1079:in `eval'
    from /Users/plundberg/.rvm/gems/jruby-1.7.26/bin/bundle:1:in `(root)'jruby-1.7.26 :003 > quit

When I switch to JRuby 1.7.25, all is fine:

$ rvm use 1.7.25
Using /Users/plundberg/.rvm/gems/jruby-1.7.25
$ bundle exec irb
jruby-1.7.25 :001 > require 'sequel'
 => true
jruby-1.7.25 :002 > Sequel.extension :core_extensions
 => [:core_extensions]
jruby-1.7.25 :003 > quit

The failing code (https://github.com/jeremyevans/sequel/blob/master/lib/sequel/core.rb#L139) looks like this:

  def self.extension(*extensions)
    extensions.each{|e| Kernel.require "sequel/extensions/#{e}"}
end

Any ideas or suggestions on what change in 1.7.26 could have been causing this?

/cc @olleolleolle, @sdahlbac
@olleolleolle
Copy link
Member

A workaround:

require 'sequel'
module Sequel
  def self.extension(*extensions)
    extensions.each { |e| Kernel.send(:require, "sequel/extensions/#{e}") } # PATCH: avoid Kernel.require being private
  end
end
Sequel.extension :core_extensions

@jayjlawrence jayjlawrence mentioned this issue Sep 14, 2016
Closed
@jayjlawrence
Copy link

There is a similar issue with Bundler 1.13 and 1.13.1. The issue is also present in jRuby 1.7.24. I haven't checked other versions. See the referenced issue above.

@olleolleolle
Copy link
Member

@perlun Perhaps your jruby-1.7.25 has bundler v1.12.5 installed?

@olleolleolle
Copy link
Member

olleolleolle commented Sep 14, 2016 •
edited

The changed Bundler code here, in the #reverse_rubygems_kernel_mixin method - shown next to its older implementation:
rubygems/bundler@c144eb6

@headius
Copy link
Member

headius commented Sep 14, 2016

This is very strange. Kernel.require appears to be public and callable in all recent JRuby versions I tried.

@headius
Copy link
Member

headius commented Sep 14, 2016 •
edited

Ok, so the problem is definitely from #reverse_rubygems_kernel_mixin, which attempts to restore the original require over the top of the patched version from RubyGems. In this case, it yanks gem_original_require out as an unbound method and redefines it as require. It basically transplants the old one back in.

rubygems/bundler@c144eb6#diff-f271319d553ee0af23e993e254a92c6cR293

In MRI, that causes visibility to be set to whatever is current for the frame, which is usually public. So the new require is public.

In JRuby, because we associate visibility with the method rather than with the method table entry, the original private visibility sticks to the method object and it remains private after transplanted. So this logic causes require to be private when it was public.

Working with bundler folks on a fix now in rubygems/bundler#4975

@headius
Copy link
Member

headius commented Sep 14, 2016

We have proposed to bundler that they modify the redefine_method logic to also send(:public, ...) after the transplanting. It will hurt no Ruby and will allow the logic to work on any JRuby version.

FWIW, recent 9k (since 9.1.3.0) does not appear to have this problem. I fixed #3869 in 9b4e97b on master but did not backport that fix to 1.7. All versions of JRuby prior to 1.7.27 and 9.1.3.0 will have this bug.

headius added a commit that referenced this issue Sep 14, 2016
@headius
Stop using WrapperMethod, since it messes up super logic.
f11ea4b 
This is for #3869 and relates to the module_function change from
and redefine it with a new visibility and implementation class.
However the impl class never passed through to the contained
method, preventing it from being used in super. This affected, for
example, module_fuction singleton methods that need to super or
methods transplanted using defined_method with a Method instance.
The new logic always tries to dup the target method so it can
be truly populated with the altered fields. This change fixed

The previous commit, using cloning instead of construction for
IR methods, works around the fact that there's no semi-transparent
WrapperMethod to delegate its serial number to the wrapped method.
Since in that case and in this one, the method's serial number
was expected to be the same after duplication, the clone
technique seems acceptable.

jruby-1_7: Fixes #4155
@headius headius added this to the JRuby 1.7.27 milestone Sep 14, 2016
headius added a commit that referenced this issue Sep 14, 2016
@headius
Add a spec for #4155.
3b556f6
headius added a commit that referenced this issue Sep 14, 2016
@headius
define_method should modify the visibility of an UnboundMethod.
70d1fc2 
See #4155 and rubygems/bundler#4975.
@headius
Copy link
Member

headius commented Sep 14, 2016

Thanks for the footwork on this one, @olleolleolle and @jayjlawrence, you made it easy to track down.

@headius headius closed this as completed Sep 14, 2016
@perlun
Copy link
Contributor Author

perlun commented Sep 15, 2016

Thanks a lot for sorting this out @headius, @olleolleolle & @jayjlawrence - I'm happy to have been wrong this time (it wasn't really a regression in JRuby). Even happier that it has been fixed now. 👍

eregon pushed a commit to ruby/spec that referenced this issue Sep 27, 2016
@headius @eregon
define_method should modify the visibility of an UnboundMethod.
c37bf53 
See jruby/jruby#4155 and rubygems/bundler#4975.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
core JRuby 1.7.x needs tests
Projects
None yet
Milestone
JRuby 1.7.27
Development

No branches or pull requests
4 participants
@olleolleolle @headius @perlun @jayjlawrence