On Tue, Jan 31, 2012 at 12:15:17AM -0800, Darren Duncan wrote: > You said, "provided said call does not throw an exception". > > So does that mean you have also promoted some runtime exceptions to > compile time exceptions, or inlined the exception toss? > > That would be useful. You don't have to wait until runtime to know > that the code is going to die unconditionally. > > So constant folding can then work like this: > > 1. If no exception will be thrown when using the constant arguments > you have, replace the call with the result value as per usual. > > 2. If an exception will certainly be thrown when using the constant > arguments, then either: > > 2a. Where you would otherwise have put the folded result, place a "fail"/etc. > > 2b. Throw an exception at compile time. > > Generally speaking, the more kinds of errors you can catch at > compile time, the more reliable your program is.
Suprisingly I think this is a good idea. Implemented. (2b) -Stefan
signature.asc
Description: Digital signature
