bug#12119: 24.1.50; symbol-macrolet regresssion

[Helmut Eller]

On Sun, Aug 05 2012, Stefan Monnier wrote:

>> Also, in Emacs 22 the macroepansion (via cl-macroexpand-all) 
>> produces something like
>
>> (let ((outer '42))
>>   (let ((outer 'inner))
>>     (progn
>>       (or
>>        (eq outer 'inner)
>>        (signal 'cl-assertion-failed
>>                (list
>>                 '(eq x 'inner))))
>>       nil)))
>
> While the above expansion returns the right value, it is not correct.
> The right expansion would be
>
>    (let ((outer '42))
>      (let ((x 'inner))
>        (progn
>          (or
>           (eq x 'inner)
>           (signal 'cl-assertion-failed
>                   (list
>                    '(eq x 'inner))))
>          nil)))
>
>> while in Emacs 24 it looks like 
>
>> (let ((outer '42))
>>   (progn
>>     (let ((x 'inner))
>>       (progn
>>      (or
>>       (eq outer 'inner)
>>       (signal 'cl-assertion-failed
>>               (list
>>                '(eq x 'inner))))
>>      nil))))
>
>> Such a change should be documented.
>
> That's an accident, so it shouldn't be documented.  The new code is no
> better than the old one, and arguably slightly worse since it fails in
> your test while the old code worked.
> I'm not yet sure how we can fix it properly, tho.

Well, it depends on the definition of "correct".  The old version seems
to do what the documentation says, i.e., let binding the variable is
treated like letf.  What you call "correct" may be closer to what ANSI
CL does, but such a change should be documented.

Either way, I don't see the point in breaking existing code.

Helmut