Comments on: SICP 3.4 http://eli.thegreenplace.net/2007/10/26/sicp-334/ Eli Bendersky's personal website Fri, 21 Nov 2008 16:52:44 +0000 http://wordpress.org/?v=2.6.2 By: Vlad http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-133608 Vlad Sat, 11 Oct 2008 15:22:53 +0000 http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-133608 About answer to 3.41 Scheme has unbound integers, that can occupate more bytes that machine word of host computer. That means that writes to memory will not be atomic and balance accessor can return "garbage"(like 8 byte number on 32-bit host that has first 4 bytes updated by set! operation in withdraw or deposit procedures and last 4 bytes from old value). About answer to 3.41

Scheme has unbound integers, that can occupate more bytes that machine word of host computer. That means that writes to memory will not be atomic and balance accessor can return “garbage”(like 8 byte number on 32-bit host that has first 4 bytes updated by set! operation in withdraw or deposit procedures and last 4 bytes from old value).

]]> By: eliben http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-125300 eliben Thu, 07 Aug 2008 17:29:05 +0000 http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-125300 nobody, Thanks - I've fixed both 3.47a and 3.47b nobody,
Thanks - I’ve fixed both 3.47a and 3.47b

]]> By: nobody http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-125143 nobody Mon, 04 Aug 2008 23:54:25 +0000 http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-125143 I think there is a bug in 3.47a. The comparison (> count 0) should be done while mutex is acquired, too. With current implementation the following is possible: * count = 1 * thread T1 finds out that count > 0, then T1 is swapped out * T2 finds out that count > 0, acquires the mutex, updates count to 0, releases mutex * T1 acquires the mutex, updates count to -1... oops I think there is a bug in 3.47a. The comparison (> count 0) should be done while mutex is acquired, too. With current implementation the following is possible:
* count = 1
* thread T1 finds out that count > 0, then T1 is swapped out
* T2 finds out that count > 0, acquires the mutex, updates count to 0, releases mutex
* T1 acquires the mutex, updates count to -1… oops

]]> By: mats http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-115503 mats Wed, 12 Mar 2008 04:11:41 +0000 http://eli.thegreenplace.net/2007/10/26/sicp-334/#comment-115503 thanks, this helped me. thanks, this helped me.

]]>