My intuition on what the names implied, as well as some observations, tells me that BEGIN simply starts a new state and destroys the whole state stack on hold, while yy_push_state push a new state into the state stack, so that later we can use yy_pop_state to get the caller state back.
Is my intuition correct?
It turns out the behavior is well-documented at here, I somehow ignored it when searching.
My assumptions stated in question is half correct but half wrong.
I am correct in that yy_push_state(int new_state) pushes the currently activated state into stack and activates the new_state. yy_pop_state() pop the state at the top of the stack then activate it. See the toy GDB test below:
(gdb) p yy_push_state(0)
$30 = void
(gdb) p yy_push_state(1)
$31 = void
(gdb) p yy_push_state(2)
$32 = void
(gdb) p yy_push_state(3)
$33 = void
(gdb) p yy_top_state()
$34 = 2
(gdb) p yy_pop_state()
$36 = void
(gdb) p yy_top_state()
$37 = 1
I am wrong in that BEGIN has nothing to do with stack, it simply activate its argument. The stack is not affected by BEGIN whatsoever.
#define BEGIN (yy_start) = 1 + 2 *