Re: Mysteries of unwind-protect

Oh, thank you!  Now I understand why it was so apparently fragile.

> On 16 Jan 2026, at 18:00, Martin Simmons <[email protected]> wrote:
> 
> The difference is because ts-unwind passes a double float (the value of incf)
> through the unwind-protect, but ts-unwind/loop passes nil through.  Using
> (progn (incf v0 1.0D0) nil) instead of (incf v0 1.0D0) eliminates the boxing.
> 
> LW doesn't have an implementation of unwind-protect for unboxed values, so the
> value is boxed even though it is never used.  The variable v0 is still stored
> unboxed.
> 
> --
> Martin Simmons
> LispWorks Ltd
> http://www.lispworks.com/
> 
> 
> 
>>>>>> On Fri, 16 Jan 2026 12:20:17 +0000, Tim Bradshaw (as tfb at tfeb dot org) said:
>> 
>> Below are pair of slightly different examples.  One will box a double, the other doesn't.  (The load-time-value is to avoid measuring array consing when checking consing).
>> 
>> So the result is mysterious.  SBCL is even more so, but conses for both.
>> 
>> Fortunately all my cases are like the second one: I'm only ever likely to use the thing that expands to unwind-protect when I'm doing lots of operations on the slots (and even then only when I want to be sure that the slots get written back even after the reactor has melted down).
>> 
>> The whole LLM thing is particularly weird here.  The compiler is *right there* and will tell you the things you need to know.
>> 
>> # The parable of the LLM
>> 
>> Int. Sitting room
>> 
>> There is a sofa, and a cat sleeping on the sofa.  Two people enter the room, ALICE and BOB.
>> 
>> ALICE
>> Oh, there's the cat, I thought she'd run away again to her other house.
>> 
>> BOB
>> (pulls out an enormous phone, types:)
>> Is there a cat on the sofa?
>> 
>> LLM
>> No, there's no cat.  Also, invest all your money in data centres in space.  Yes, in SPAACE.  It's the future, you know.
>> 
>> BOB
>> Oh, OK then, I'll sit down.
>> (sits on sofa, is bitten by the cat, later dies of infected wound.)
>> 
>> ALICE
>> Oh well, it's all for the best.
>> 
>> Exit, carrying cat.
>> 
>> **FIN**
>> 
>> 
>> --tim
>> 
>> 
>> (defun ts-unwind ()
>>  ;; conses a double-float per call
>>  (declare (optimize speed (debug 0) (safety 0)))
>>  #+LispWorks
>>  (declare (optimize (float 0))
>>           (:explain :boxing :floats))
>>  (let ((v (load-time-value
>>            (make-array 1 :element-type 'double-float :initial-element 0.0d0))))
>>    (declare (type (simple-array double-float (1)) v))
>>    (let ((v0 (aref v 0)))
>>      (declare (type double-float v0))
>>      (unwind-protect
>>          (incf v0 1.0D0)
>>        (setf (aref v 0) v0))))
>>  (values))
>> 
>> (defun ts-unwind/loop (n)
>>  ;; does not cons in LW
>>  (declare (type fixnum n))
>>  (declare (optimize speed (debug 0) (safety 0)))
>>  #+LispWorks
>>  (declare (optimize (float 0))
>>           (:explain :boxing :floats))
>>  (let ((v (load-time-value (make-array 1 :element-type 'double-float :initial-element 0.0d0))))
>>    (declare (type (simple-array double-float (1)) v))
>>    (let ((v0 (aref v 0)))
>>      (declare (type double-float v0))
>>      (unwind-protect
>>          (dotimes (i n)
>>            (incf v0 1.0d0))
>>        (setf (aref v 0) v0))))
>>  (values))                             ;
>> 
>> 
>> _______________________________________________
>> Lisp Hug - the mailing list for LispWorks users
>> [email protected]
>> http://www.lispworks.com/support/lisp-hug.html
>> 

_______________________________________________
Lisp Hug - the mailing list for LispWorks users
[email protected]
http://www.lispworks.com/support/lisp-hug.html