Merge pull request #29 from opscode/lcg/better-ext-vs-ffi-logic

add better ext-vs-ffi logic

1 files changed, 22 insertions, 2 deletions

diff --git a/lib/ffi_yajl.rb b/lib/ffi_yajl.rb
index ce099e0..7f39189 100644
--- a/lib/ffi_yajl.rb
+++ b/lib/ffi_yajl.rb
@@ -1,13 +1,33 @@
 
+#
+# Precedence:
+#
+# - The FORCE_FFI_YAJL env var takes precedence over everything else, the user
+#   theoretically knows best
+# - Java always gets ffi because jruby only supports ffi
+# - There is a conflict between loading libyajl 1.x and 2.x in the same VM
+#   process (on a fundamental basis, simply guru medidate about how the
+#   c-symbols work if you load both libs).  For some reason the ffi interface
+#   seems to work fine sometimes (i'm not sure how) so we fall back to that--
+#   this is much more likely to be converted into a raise than to have the warn
+#   dropped, so don't bother asking for that.
+# - Then we try the c-ext and rescue into ffi that fails
+#
 if ENV['FORCE_FFI_YAJL'] == "ext"
   require 'ffi_yajl/ext'
-elsif ENV['FORCE_FFI_YAJL'] == "ffi" || defined?(Yajl)
-  # - c-extension segfaults on ruby 1.8.7
+elsif ENV['FORCE_FFI_YAJL'] == "ffi"
+  require 'ffi_yajl/ffi'
+elsif RUBY_PLATFORM == "java"
+  require 'ffi_yajl/ffi'
+elsif defined?(Yajl)
+  warn "the ffi-yajl and yajl-ruby gems have incompatible C libyajl libs and should not be loaded in the same Ruby VM"
+  warn "falling back to ffi which might work (or might not, no promises)"
   require 'ffi_yajl/ffi'
 else
   begin
     require 'ffi_yajl/ext'
   rescue LoadError
+    warn "failed to load the ffi-yajl c-extension, falling back to ffi interface"
     require 'ffi_yajl/ffi'
   end
 end