Ruby 2.2 got rid of rb_thread_blocking_region?

[FooBarWidget]

It is reported that, when running Passenger under Ruby 2.2, the following error occurs:

passenger_native_support.so: undefined symbol: rb_thread_blocking_region

There have probably been large changes with regard to thread handling. This needs to be investigated.

[peterc]

I just had a look at the main Ruby repo and this might help: https://github.com/ruby/ruby/blob/1c87fee3166c5208e92f1b7779507dd52c3087e1/NEWS

A variety of deprecated C APIs have been removed in Ruby 2.2. More info at https://bugs.ruby-lang.org/issues/9502 - a variety of rb_thread_blocking_* things have been replaced by the "rb_thread_call_without_gvl family" whatever that is ;-)

I don't know if this also helps but Googling some of the terms involved brought up some code where another project has updated their own C code to fit both the old and new style APIs - just in case it's useful in your own research: https://www.mail-archive.com/clogger@librelist.org/msg00022.html

[FooBarWidget]

This is odd. I can't reproduce this with ruby-head 2.2.0dev, 2014-12-12.

[FooBarWidget]

Never mind that last comment. :)

[xenda]

I was just testing Passenger 4.0.55 and Passenger 4.0.56 with Ruby 2.2.0p0 and have find out the problem still exists:

App 22697 stderr: Passenger RackApp: /home/app/app: symbol lookup error: /usr/local/lib/ruby/gems/2.2.0/gems/passenger-4.0.55/buildout/ruby/ruby-2.2.0-x86-linux/passenger_native_support.so: undefined symbol: rb_thread_blocking_region
2014/12/25 12:05:22 [error] 22688#0: *25 upstream prematurely closed connection while reading response header from upstream, client: 72.46.130.42, server: XXXX, request: "GET / HTTP/1.1", upstream: "passenger:/tmp/passenger.1.0.22668/generation-0/request:"

Any ideas if we are doing something wrong?

[FooBarWidget]

Maybe something went wrong during compilation. Could you run this and tell me what you see?

sudo rm -rf /usr/local/lib/ruby/gems/2.2.0/gems/passenger-4.0.55/buildout/ruby/ruby-2.2.0-x86-linux
passenger-config build-native-support

[xenda]

Tried compiling it again, removing it and creating it with the build-native-support.

Still failing, but I see the log mentions that symbol:

# /usr/local/bin/ruby /usr/local/lib/ruby/gems/2.2.0/gems/passenger-4.0.56/ext/ruby/extconf.rb
checking for alloca.h... yes
checking for ruby/version.h... yes
checking for ruby/io.h... yes
checking for ruby/thread.h... yes
checking for ruby_version... no
checking for rb_thread_io_blocking_region()... no
checking for rb_thread_call_without_gvl()... no
creating Makefile
# make clean && make
compiling /usr/local/lib/ruby/gems/2.2.0/gems/passenger-4.0.56/ext/ruby/passenger_native_support.c
/usr/local/lib/ruby/gems/2.2.0/gems/passenger-4.0.56/ext/ruby/passenger_native_support.c: In function ‘f_generic_writev’:
/usr/local/lib/ruby/gems/2.2.0/gems/passenger-4.0.56/ext/ruby/passenger_native_support.c:341:6: warning: implicit declaration of function ‘rb_thread_blocking_region’ [-Wimplicit-function-declaration]
      ret = (int) rb_thread_blocking_region(writev_wrapper,
      ^
linking shared-object passenger_native_support.so

[nanaya]

the referenced commits (0ba44a9, bf8de31) fix usage of rb_thread_io_blocking_region, not rb_thread_blocking_region.

[JamesHarrison]

Can confirm I'm also seeing this with 4.0.56 and 5.0p2.

App 31609 stderr: Passenger RackApp: /home/appname/current: symbol lookup error: /home/appname/.passenger-enterprise/native_support/4.0.56/ruby-2.2.0-x86_64-linux/passenger_native_support.so: undefined symbol: rb_thread_blocking_region

... and 5.x:

App 3192 stderr: Passenger RubyApp: /home/appname/releases/20141226021229: symbol lookup error: /home/appname/.rbenv/versions/2.2.0/lib/ruby/gems/2.2.0/gems/passenger-enterprise-server-5.0.0.beta2/buildout/ruby/ruby-2.2.0-x86_64-linux/passenger_native_support.so: undefined symbol: rb_thread_blocking_region

The extensions all seem to compile cleanly, though. Tried recompiling without any change.

Edit: @edogawaconan quite right, hadn't spotted. Does that want a new issue creating to track?

[nanaya]

looking at the deprecation list, rb_thread_io_blocking_region isn't even in it. Hitting wrong target?

(yeah, I'm also getting the same error as everyone else here)

[FooBarWidget]

No, the committed code is correct. We check for rb_thread_call_without_gvl() and use that instead of rb_thread_(io_)blocking_region. The problem is that the rb_thread_call_without_gvl() check fails for some reason, as can be seen in @xenda 's logs.

@xenda Is there an mkmf.log in the directory? If so could you attach it?