bug#19954: python.el: more consistent sexp navigation

[Carlos Pita]

X-Debbugs-CC: fgallina@gnu.org

If this is a feature I recognize I don't understand the rationally
behind it:

        [...]
            (and (not forward-p)
                 (eq (syntax-class (syntax-after (1- (point))))
                     (car (string-to-syntax ")")))))
        ;; Inside a paren or looking at it, lisp knows what to do.

Say * is the point. The inconsistency I find is that C-M-left will do
very different things while at the end of different lines:

A)
from sklearn.cross_validation import KFold*
-->
*from sklearn.cross_validation import KFold

B)
n = len(train.y)*
-->
n = len*(train.y)

I think the intention is to nav at the "statement/block level" when the point
is at the end of the line. The difference of behaviour between A and B
can't be reconciled at any level: sexp, list, statement, block. My
expectation would be:

B')
n = len(train.y)*
-->
*n = len(train.y)

B'')
n = len(train.y*)
-->
n = len(train.*y)

Notice that this is even more conspicuous at the end of a block:

C)
for x in range(0, 10):
    x = 2
    print("hello")*
-->
for x in range(0, 10):
    x = 2
    print*("hello")

D)
for x in range(0, 10):
    print("hello")
    x = 2*
-->
*for x in range(0, 10):
    print("hello")
    x = 2

I vote for removing the (syntax-after (1- (point)) special case as IMO
it only adds confussion to the already complex nav rules.

Cheers
--
Carlos