Complete Verifiable Example of rgeo/rgeo-proj4#39

You need to have proj installed for this to work (ex: brew install proj).

./run # reproduce the error
./run fix # reproduce a fix

The bug happens when we are forking process in Ruby and not using Proj4's multithreaded API. I'm sure that we could pin the bug further, and I'd be glad to know how if someone is doing it (buonomo.ulysse[at]gmail.com or a comment).

What's next

I didn't have rgeo-proj4 offline so I couldn't code a fix. We should use the proper proj_context* API rather than using the default context. I also tested further and it seems like having a local static PJ_CONTEXT* in the C extension file, and set it in the Init_extension method should be sufficient. Even if we fork afterward. But of course this would be better with a test reproducing the bug (e.g. non-regression test).

How did I get there

From Pe-co's reproduction and using ruby's TracePoint API to know where were all the c_calls where coming from. I could progressively remove parts of the example to understand that nor RGeo nor Rails had anything to do with it. And in the end, in the proj4_create function in the rgeo-proj4 project I had to switch to C, ending with this example.