Patch: Fix for PR java/4766

[Per Bothner]

Tom Tromey wrote:

> 	For PR java/4766:
> 	* jcf-write.c (generate_bytecode_insns) [TRY_FINALLY_EXPR]: Handle
 	case where `finally' clause can't complete normally.


This is OK.

Two concerns:

(1) This patch is basically an optimization.  If we just added a
if (CAN_COMPLETE_NORMALLY (finally)) before the 'ret', would that
work (though less efficiently)?  Otherwise, I'm wondering if we're
just papering over some more subtle bug.

(2) Does the compiler reject now reject the invalid bytecode (as
generated by the old compiler) with a reasonable error message?
Otherwise I wouldn't consider the PR fully fixed.
-- 
	--Per Bothner
per@bothner.com   http://www.bothner.com/per/