[Python-Dev] [SPAM: 3.000] [issue11682] PEP 380 reference implementation for 3.3

[Nick Coghlan]

On Wed, Nov 9, 2011 at 8:59 PM, Greg Ewing <greg.ewing at canterbury.ac.nz> wrote:
> Nick Coghlan wrote:
>> I'll add a new test to ensure "yield from x" requires parentheses whenever
>> "yield x" requires them (and fix the Grammar file on the implementation
>> branch
>> accordingly).
>
> Wait a minute, there's nothing in the PEP as accepted that
> mentions any such restriction.

It doesn't need to be mentioned in the PEP - it's inherited due to the
fact that the PEP builds on top of yield expressions. There is no
excuse for the two expressions being inconsistent in their parenthesis
requirements when they are so similar syntactically.

For ordinary yield expressions, "f(yield x, y)" has to be disallowed
since it is genuinely ambiguous (does it mean "f(yield (x, y))" or
"f((yield x), y)"?), and it then becomes something of a pain to allow
any yield argument other than an unparenthesised tuple when a yield
expression is provided as the sole argument to a call.

So, from my point of view, it is absolutely a requirement that 'yield
from' expressions be parenthesised in the multiple argument case. The
previous grammar allowed confusing constructs like "f(yield from x,
y)" (interpreting it as "f((yield from x), y) while still issuing a
syntax error (as expected) for "yield from x, y".

You *could* try to make a case that we should allow "f(yield from x)",
but it isn't then clear to the casual reader why that's OK and
"f(yield x)" gets disallowed. And that brings us back to the
complexity of creating a special yield expression variant that's
allowed as the sole argument to a function call without additional
parentheses, but without introducing ambiguity into the grammar.

It's simpler and cleaner to just make the rules for the two constructs
identical - if it's a subexpression, you have to add the extra
parentheses, if it's a statement or the sole expression on the RHS of
an assignment, you don't.

That said, it's likely *possible* to make those parentheses optional
in the single argument call case (as we do for generator expressions),
but do you really want to hold up the PEP 380 implementation for a
minor detail like that? Besides, since it would affect the grammar
definition for ordinary yield expressions as well, it might even need
a new PEP.

Cheers,
Nick.

-- 
Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia