You also need to know how tokens are produced by lexical analysis. It could have produced:

  'num1' '+' '++' 'num2'

In fact it produces

  'num1' '++' '+' 'num2'

so it's not about evaluate from left-to-right.