Support Ruby's bigdecimal library

Is there an existing test suite for Ruby's BigDecimal extension anywhere? I've searched but could not find one.

If there's not in either rubicon or in Ruby's source, then I don't know where else there'd be one. You might ask on ruby-core or ruby-talk mailing lists, but it's highly possible that there's no test suite out there.

I committed a partial implementation from Ola, but many of the methods are just stubbed out. There's more work to do, but non-bigdecimal Rails 1.2 tests will run now.

The current bigdecimal doesn't provide a global function, i.e., BigDecimal(string): #new has to be explicit. However, bigdecimal/util.rb uses the non-new syntax.

Used by AR for decimal type columns.

And doesn't compare well to floats. I tried looking into this but got stuck: haven't been able to figure out how jruby implements this kind of comparison. Doesn't call <=>? Doesn't seem to or I'm missing something.

I wonder if the incomplete functions in BD should raise something rather than return bad data? It'd save time to get a definite idea rather than trying to track down latent wrong results?

It looks like the non-explicit new is now supported, although like the explicit new, it's more permissive than its MRI equivalent: not only are strings accepted, but so are Fixnums, Floats, other BigDecimals, and possibly other types. In both the explicit and implicit constructions, MRI seems to want to #to_str the argument, and if it can't it raises a TypeError. I'm not sure why MRI is so persnickety in this regard, but it is consistent with the bigdecimal documentation.

Attached are some patches to make JRuby's BigDecimal reject first arguments not responding to #to_str, as well as tests for the some of the problems documented in this case.

JRUBY-341 documents BigDecimal's problems with integers, and here we have mention of problems with floats. If those two issues are related, might the float issue be folded into JRUBY-341 and this issue closed?