go version devel +746441f97f Wed Apr 12 18:16:41 2017 +0000 linux/amd64

Consider the following:

package p

var _ = float32(1) * 1e200
var _ = float64(1) * 1e500

var _ = complex64(1) * 1e200
var _ = complex128(1) * 1e500

on tip (1.8 too) we get a double error on the first two lines (but not on the last two):

./prova.go:3:16: constant Inf overflows float32
./prova.go:3:20: constant 1e+200 overflows float32
./prova.go:4:16: constant Inf overflows float64
./prova.go:4:20: constant 1e+500 overflows float64

./prova.go:5:22: constant 1e+200 overflows complex64
./prova.go:6:23: constant 1e+500 overflows complex128

The lines complaining about Inf overflow should go (they're confusing, there's no Inf in the code).

It should be enough to avoid overwriting the original Node's Val in case of overflow in truncfltlit( ) (like we already do in the recently committed trunccmplxlit( )).