Philip Kaludercic wrote:
>> +(cl-defstruct (stream (:constructor stream--make-stream)
>> +                      (:predicate streamp)
>> +                      :named)
>> +
>> +  "A lazily evaluated sequence, compatible with the `seq' library's functions."
>> +
>> +  (evaluated--internal
>> +   nil
>> +   :type boolean
>> +   :documentation "Whether the head and tail of the stream are accessible.
>> +
>> +This value is set to t via the function `stream--force' after it
>> +calls the updater function.")
>> +
>> +  (first--internal
>> +   nil
>> +   :type (or t null)
> 
> Isn't this type just t?  Not a proof, but this doesn't signal:

Yes. I have changed it to just `t`.

>> +(defun stream--force (stream)
>> +  "Evaluate and return the STREAM.
>> +
>> +If the output of the updater function is nil, then STREAM is
>> +marked as empty.  Otherwise, the output of the updater function
>> +is used to set the head and the tail of the stream."
>> +  ;; Check explicitly so that we can avoid checking
>> +  ;; in accessors by setting safety to 0 via `cl-declaim'.
>> +  (unless (streamp stream)
>> +    (signal 'wrong-type-argument (list 'stream stream)))
> 
> If you are already using cl-lib, you could also make use of `cl-check-type'.

Done.

>> +  (if (stream-evaluated--internal stream)
>> +      stream
>> +    (pcase (funcall (stream-updater--internal stream))
>> +      (`(,head . ,tail)
>> +       (setf (stream-first--internal stream) head
>> +             (stream-rest--internal stream) tail))
>> +      ((pred null)
>> +       (setf (stream-empty--internal stream) t))
>> +      (bad-output
>> +       (error "Bad output from stream updater: %s"
>> +              bad-output)))
>> +    (setf (stream-evaluated--internal stream) t)
>> +    stream))
>>
>>   (defmacro stream-make (&rest body)
>>     "Return a stream built from BODY.
>> -BODY must return nil or a cons cell whose cdr is itself a
>> -stream."
>> -  (declare (debug t))
>> -  `(cons ',stream--fresh-identifier (lambda () ,@body)))
>>
>> -(defun stream--force (stream)
> 
> Did you change the order of the definitions?

Yes. I brought the fundamental features to the top of the file. In the 
existing version, things are defined after they are used. I know that is 
not an error, but it had me jumping around more to see how it worked.

>> +
>> +(defconst stream-empty
>> +  (stream--make-stream :evaluated--internal t
>> +                       :first--internal nil
>> +                       :rest--internal nil
>> +                       :empty--internal t
>> +                       :updater--internal nil)
>> +  "The empty stream.")
>> +
>> +(defun stream-empty ()
>> +  "Return the empty stream."
>> +  stream-empty)
> 
> This definition also appears unchanged?  Please try to keep the diff as
> small as possible.

OK. I have broken it into several files, which I've attached. How does 
it look now?