On Wed, 04 Nov 2009 21:39:02 -0500, Vassil Nikolov <vniko...@pobox.com> said:
> On Wed, 4 Nov 2009 18:17:12 -0800 (PST), Spiros Bousbouras <spi...@gmail.com> said: >> In http://www.lispworks.com/documentation/HyperSpec/Body/m_rotate.htm >> we read: "In the form (rotatef place1 place2 ... placen), the values >> in place1 through placen are read and written". Does it mean "saved" >> instead of "written" ? > No; here "write" is used in a sense similar to the first meaning of > "write" given in the Glossary.
Or maybe I misunderstood the question (wasn't it at least a little ambiguous, though?). Perhaps a better answer would be that the fact that the values read are _saved_, if that means "assigned to temporary variables inside ROTATEF", is an internal implementation detail, while the fact that the values of the places are read, and then values (which form a permutation of the values read) are _written_ into those places, is part of ROTATEF's specification.
---Vassil.
-- "Even when the muse is posting on Usenet, Alexander Sergeevich?"
The text makes no sense. Taken as is, it implies that the values are stored twice. It looks like the result of careless edit; as if originally, the text had said that all of the values are read and written, and then went on to explain what exactly happens.
You can fix it simply by deleting the word "then"; it is the "then" separating the "written" and "stored" which makes it look like they are distinct, sequenced events, rather than independent remarks on the same situation.
Also, notably abesent is a ``Side Effects:'' paragraph, but this inconsistency is repeated elsewhere in the document, like SETF/PSETF.
On Thu, 5 Nov 2009 04:41:02 +0000 (UTC), Kaz Kylheku <kkylh...@gmail.com> said:
> On 2009-11-05, Spiros Bousbouras <spi...@gmail.com> wrote: >> In http://www.lispworks.com/documentation/HyperSpec/Body/m_rotate.htm >> we read: "In the form (rotatef place1 place2 ... placen), the values >> in place1 through placen are read and written". Does it mean "saved" >> instead of "written" ? > The text makes no sense. Taken as is, it implies that the values are > stored twice. It looks like the result of careless edit; as if > originally, the text had said that all of the values are read and written, and > then went on to explain what exactly happens. > You can fix it simply by deleting the word "then"; it is the "then" separating > the "written" and "stored" which makes it look like they are distinct, > sequenced events, rather than independent remarks on the same situation.
It seems I was more wrong than I thought earlier in my preceding responses---quite wrong, in fact, and it indeed is proper that it should be "saved" there (have I been just not noticing that "then"...). In other words, it is specified that _all_ values are obtained _before_ _any_ place is changed, which would matter in case the value of one of the places depends on the value of another. That is independent of any side effects which might occur while accessing a place.
In other words, assuming that no side effects are taking place while accessing P1 or P2, and assuming that fresh symbols are used as needed,
Now I can even actually recognize this as being related to the notorious inability of Algol's call-by-name to support a universal swapping operation, with the essence of the counterexample being
(rotatef n (aref a n))
---note how there are no side effects when accessing either N or (AREF A N) and how (i)'s way from above would lead to
(let (t1) (setf t1 n) (setf n (aref a n)) (setf (aref a n) t1))
doing the wrong thing.
---Vassil.
-- "Even when the muse is posting on Usenet, Alexander Sergeevich?"
On Thu, 05 Nov 2009 00:25:50 -0500, Vassil Nikolov <vniko...@pobox.com> said:
> ... > but [(ROTATEF P1 P2)] must be executed as if it expands into > ;; (ii) > (let (t1 t2) > (setf t1 p1) > (setf t2 p2) > (setf p1 t2) > (setf p2 t1))
...with the mechanics of setf expanders _having been applied_, so that there is a "safe fix" on the storage locations for P1 and P2 before any updates are made. That proviso is significant because without it
(let (t1 t2) (setf t1 n) (setf t2 (aref a n)) (setf n t2) (setf (aref a n) t1))
loses, too. (Algol's problem was that it did not have a mechanism equivalent to Common Lisp's setf expanders.)
I hope it takes me no more than a finite number of posts to catch all my mistakes in this thread...
---Vassil.
-- "Even when the muse is posting on Usenet, Alexander Sergeevich?"
|
