On Sun, Feb 13, 2005 at 10:33:06PM +0100, Tomasz Zielonka wrote:
On Sun, Feb 13, 2005 at 10:25:49PM +0100, Remi Turk wrote:
BTW, I have an implementation of STM based entirely on old concurrency primitives, which means that it will work in older GHC and probably in other Haskell compilers. I am going to put it on my web site, when I get one.
Cool :) Is it actually race/deadlock/othergeneralnastiness-free?
It should be, but there may be bugs of course. I know there are some possible space leaks, but that's also fixable.
(as the paper claims that e.g. mergeIO :: [IO a] -> IO a is unimplementable in anything built on mutexes and condition variables.)
My STM monad is not IO, it has the same restrictions as STM in the paper. The paper doesn't claim you can't implement mergeSTM :: [STM a] -> STM a mergeSTM = msum
Ugh, I should've read what I copied... *nitpick* in the paper merge is defined using foldr1, which errors when given an empty list, as opposed to msum, which merely retries.
Also, the STM in GHC 6.4 is written in C. Do you think that Haskell's IO monad lacks some things needed to write this thing? I can tell you there are some problems with types, but they can be solved in a more or less standard way.
Best regards Tomasz
I don't, though I'm probably not qualified to vote ;) Groeten, Remi P.S. And don't forget to post a link when you put it online. -- Nobody can be exactly like me. Even I have trouble doing it.