Properly handle missing binary libraries.

[eendeego]

After removing a binary library dependency from my project (in this case: libffi), bundler started to crash on simple commands as:

rake -T

Error message:

rake aborted!
undefined method `gsub' for nil:NilClass
...

Running rake -T --trace yields:

rake aborted!
undefined method `gsub' for nil:NilClass
/Users/sample/.rvm/gems/ruby-1.9.3-p194@global/gems/bundler-1.1.4/lib/bundler/runtime.rb:77:in `rescue in rescue in block in require'
/Users/sample/.rvm/gems/ruby-1.9.3-p194@global/gems/bundler-1.1.4/lib/bundler/runtime.rb:72:in `rescue in block in require'
/Users/sample/.rvm/gems/ruby-1.9.3-p194@global/gems/bundler-1.1.4/lib/bundler/runtime.rb:62:in `block in require'
/Users/sample/.rvm/gems/ruby-1.9.3-p194@global/gems/bundler-1.1.4/lib/bundler/runtime.rb:55:in `each'
/Users/sample/.rvm/gems/ruby-1.9.3-p194@global/gems/bundler-1.1.4/lib/bundler/runtime.rb:55:in `require'
...

After applying this patch, running rake -T yields:

rake aborted!
dlopen(/Users/sample/.rvm/gems/ruby-1.9.3-p194@gemset/gems/ffi-1.0.11/lib/ffi_c.bundle, 9): Library not loaded: /opt/local/lib/libffi.5.dylib
  Referenced from: /Users/sample/.rvm/gems/ruby-1.9.3-p194@gemset/gems/ffi-1.0.11/lib/ffi_c.bundle
  Reason: image not found - /Users/sample/.rvm/gems/ruby-1.9.3-p194@gemset/gems/ffi-1.0.11/lib/ffi_c.bundle

I couldn't find tests for these kinds of errors, but if they exist and someone can give some directions, I'll add tests to the patch.

[travisbot]

This pull request passes (merged fd521a2 into 17e7f34).

[dvandersluis]

I'm running into the same sort of problem with a cannot open shared object file error. Perhaps:

raise e unless $1

[cookrn]

Here's a full issue report: https://gist.github.com/4fab89b4051b5bc0cfbb

In my case, the ffi-geos gem is the believed culprit.

[rohitarondekar]

This looks like a nice addition. I don't know about tests but it doesn't seem like there are any like this. I guess you would need to create a test lib that required ffi and pretend to load a non existing dependency and check if the correct error is raised. But that doesn't sound so good. @indirect @hone thoughts?

[eendeego]

Dunno, it looks like @carlhuda and/or the other mantainers are flooded with other more urgent issues... this has been rotting here for a while.

[indirect]

The test for this should just raise an error that looks like the FFI error that we want to catch. This is a good patch, and I'll see about getting a test written myself if one isn't forthcoming.