Floats in ruby/protobuf are not in fact equal

[shawnfrench]

The existing test is misleading:

protobuf/ruby/tests/common_tests.rb

Lines 39 to 40 in 82e83dd

	m.optional_float = 0.5
	assert_equal 0.5, m.optional_float

This change adds a new test making the difference between ruby and protobuf floats explicit.

Something else to consider is currently the assignment of a ruby float returns the float with ruby precision. Perhaps the setter should return the float with protobuf precision?

[JasonLunn]

Interesting; this test exposes a behavior gap between the JRuby and JRuby FFI variants; the FFI-based implementation (with the passing test) uses the same C-based implementation under the hood, but the legacy JRuby implementation is built on top of the Java Protobuf jar and fails.

@haberman - does Java Protobuf handle float precision differently, or is this specific to JRuby's use of it?

[haberman]

@haberman - does Java Protobuf handle float precision differently, or is this specific to JRuby's use of it?

I suspect this is specific to JRuby's use.

I suspect that JRuby is caching the Ruby object in this.fields when the user calls m.optional_float = 0.5. Then m.optional_float is returning the user's cached object, which will have the original precision since it was not actually derived from the Java message's value.