Comments on: Incoherent Ramblings http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2007%2F09%2F12%2Fincoherent-ramblings%2F%23comment-&seed_title=Incoherent+Ramblings Mon, 02 Jan 2012 08:22:24 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Dan Creswell http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2007%2F09%2F12%2Fincoherent-ramblings%2F%23comment-172&seed_title=Incoherent+Ramblings/comment-page-1/#comment-172 Dan Creswell Fri, 14 Sep 2007 08:30:22 +0000 http://www.dancres.org/blitzblog/2007/09/12/incoherent-ramblings/#comment-172 Some bits over on Reddit: http://programming.reddit.com/info/2p7d8/comments Some bits over on Reddit:

http://programming.reddit.com/info/2p7d8/comments

]]> By: Daniel Lyons http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2007%2F09%2F12%2Fincoherent-ramblings%2F%23comment-171&seed_title=Incoherent+Ramblings/comment-page-1/#comment-171 Daniel Lyons Fri, 14 Sep 2007 01:37:50 +0000 http://www.dancres.org/blitzblog/2007/09/12/incoherent-ramblings/#comment-171 Much of what we're hearing about STM comes from the surprising current energy for functional programming. Locking being a side-effect, it doesn't fit well into the functional paradigm. STM is particularly compelling because it happens to be a formalism that fits well into Haskell's monad system, which you already have to use to do anything stateful in the language. STM is certainly easier to program in Haskell than a threading/locking system, but one has to ask whether or not this is just pushing the complexity into the compiler. Though Haskell is one of a very few languages where I would actually trust the implementors to put something that difficult into the compiler and solve it once; they are, after all, already doing some incredibly hard work in there. Apart from that there is a general distaste for threading and locking. There is a famous proof by Hans Boehm that library-based threading models cannot produce bug-free executables because certain compiler optimizations which are safe and retain meaning in serial programs produce bugs in parallel ones. People are searching for something to defect to. Much of what we’re hearing about STM comes from the surprising current energy for functional programming. Locking being a side-effect, it doesn’t fit well into the functional paradigm. STM is particularly compelling because it happens to be a formalism that fits well into Haskell’s monad system, which you already have to use to do anything stateful in the language. STM is certainly easier to program in Haskell than a threading/locking system, but one has to ask whether or not this is just pushing the complexity into the compiler. Though Haskell is one of a very few languages where I would actually trust the implementors to put something that difficult into the compiler and solve it once; they are, after all, already doing some incredibly hard work in there.

Apart from that there is a general distaste for threading and locking. There is a famous proof by Hans Boehm that library-based threading models cannot produce bug-free executables because certain compiler optimizations which are safe and retain meaning in serial programs produce bugs in parallel ones. People are searching for something to defect to.

]]> By: Not SPJ http://dancres.org/feeder/?FeederAction=clicked&feed=Comments+on+Articles+%28RSS2%29&seed=http%3A%2F%2Fwww.dancres.org%2Fblitzblog%2F2007%2F09%2F12%2Fincoherent-ramblings%2F%23comment-170&seed_title=Incoherent+Ramblings/comment-page-1/#comment-170 Not SPJ Thu, 13 Sep 2007 23:32:15 +0000 http://www.dancres.org/blitzblog/2007/09/12/incoherent-ramblings/#comment-170 http://en.wikipedia.org/wiki/Software_transactional_memory#Implementation_issues STM is a silver bullet that doesn't work. It just trades off some obvious categories of bugs for even worse race conditions. Besides that, to atomically verify its log it has to do some type of locking anyway on everything you wrote AND everything you simply READ. It actually doesn't eliminate any of the usual problems, it just moves them down a level where you don't see them -- sweeping them under the rug, as it were; AND it gives you a whole new set of Heisenbugs to torture you with. http://en.wikipedia.org/wiki/Software_transactional_memory#Implementation_issues

STM is a silver bullet that doesn’t work. It just trades off some obvious categories of bugs for even worse race conditions.

Besides that, to atomically verify its log it has to do some type of locking anyway on everything you wrote AND everything you simply READ. It actually doesn’t eliminate any of the usual problems, it just moves them down a level where you don’t see them — sweeping them under the rug, as it were; AND it gives you a whole new set of Heisenbugs to torture you with.

]]>