You are correct that additions operator is produced first when you
derive. But when you parse, you do the opposite (reduction), and the
multiplications are recognized first, wich is precisely what you want.
THis need a bit of discussion, when you compare top-down and bottom up
parsing, but basically the bottom of the tree must be know correct
before the top can be.
More to the point, this is a consequence of the fact that the parse
tree produced with such a grammar will be structured to conform the
operator operand structure defined by priorities, up to idosyncrasies
of a CF grammar definition, which are ironed out in an abstract syntax
tree (AST). In other words, this gives the parse tree a shape that is close
to the shape desired for the AST (Thanks to Raphael for suggesting
comparison with the AST).
This is why such a grammatical definition is simpler to use to
generate the AST. More importantly, it was much more convenient in
older compilers and interpreters that did not use AST, as the parse
tree reflected the priority structure.
For example, the parse tree for x + x * x
is
E
/ | \
E + T
| /|\
T T * F
| | |
F F x
| |
x x
And the AST for x + x * x
is
+
/ \
x *
/ \
x x
It is derivable from the parse tree by moving all operators up one
edge on the preceding node, and then ignoring all non-terminals,
making each linear downward path a single edge. Essentially the same
shape.
This AST is the expression tree defined by the priorities, from which the meaning of the expression can be derived (by homomorphism), from which it can be evaluated or compiled.