Stefan Monnier <monnier@iro.umontreal.ca> writes:

>>   (progv (mapcar #'car bindings)
>>       (mapcar (lambda (x) (eval (cadr x))) bindings)
>>     (funcall
>>      `(lambda ()
>>         (macroexpand-all
>>          '(progn
>>             ,@body))))))
>
> Hmm?  Why this funcall+backquote+lambda business?
> [ Oh, and be careful to pass macroexpand-all-environment as second arg
>   to macroexpand-all.  ]

No reason, just copied it from `progv' and didn't simplify.  Here's the
final version:

(defmacro let-when-compile (bindings &rest body)
  "Like `let', but allows for compile time optimization.
\n(fn BINDINGS BODY)"
  (declare (indent 1) (debug let))
  (cl-progv (mapcar #'car bindings)
      (mapcar (lambda (x) (eval (cadr x))) bindings)
    (macroexpand-all
     `(progn
        ,@body)
     macroexpand-all-environment)))

>> The code in lisp-mode.el works fine, but there's still one thing on
>> which I'm not sure: what should happen if an `eval-when-compile' wrapper
>> is omitted?
>
> Same as what happens in your earlier let-opt when the `opt' is
> omitted.

So I guess it's fine. The byte compiler will issue a warning about an
unbound variable if `eval-when-compile' is omitted by mistake. I was
thinking maybe something clever could be done, like automatically
leaving a let binding to prevent an error.

Should I install the patch?

Oleh