I remember speaking with Charlie Nutter and Tom Enebo at a conference in October in Denmark -- they were writing the JRuby bytecode compiler just then (I mean, literally, they were sitting in the hotel lobby with their laptops and writing the code). They had a problem of what to do with a huge standard ruby test file that was basically one huge method that easily exceeded the 64K[*] limit and asked what do we do in Rhino with it. I told them we do nothing -- we fail to compile them. I think they ended up implementing some mechanism for automatically breaking down such a method into multiple methods, with a top-level method invoking the broken down pieces in order. I actually have no idea whether they did it in the end, but we definitely discussed it.

I know it's not trivial -- i.e. find a big enough basic block with no abrupt control flows (i.e. break from a loop or a return), preferably one that doesn't access many local variables so you needn't pass too many arguments and return values to and from, and then extract it into a separate method.

OTOH, I can understand how nobody wishes to tackle this as it doesn't occur too often, huge functions are a discouraged practice anyway, and in those rare cases when it happens, people can usually break them down by hand.

Attila.

--

[*] BTW, it's not a 65K limit - 65536 bytes would be 64K. Actually, the limit is 65535 bytes, making it a 64K - 1 byte, or 63,999K limit. This was a public service announcement from the Nitpicking Smartpants Association. :-)

On 2007.11.22., at 1:53, William Strathearn wrote: