On 12/4/11 10:34 PM, Brad Anderson wrote:
What's the plan with doing I/O and other irreversible actions during STM transactions then?Andrei One of the PyPy guys briefly answers this in the comments of this article on the subject of PyPy and removal of the GIL. http://morepypy.blogspot.com/2011/06/global-interpreter-lock-or-how-to-kill.html
Unfortunately the post and the discussion are not very informative. The approaches discussed (write() finishes a transaction and starts another one) are rather naive and fail to address even the simplest scenarios involving e.g. interleaving input and output.
Anyway, my comment is strictly on that discussion, not on the entire effort within the Python community. When we were working on D's concurrency model Bartosz pushed for a while quite strongly in favor of STM, but I predicted it won't pan out, at least not in time for us to rely on it. It was 2008 and the issues with STM were by then understood but not resolved properly. It's still too early to make a verdict one way or another. There's a steady stream of related publications coming. It will be interesting to see what happens.
Andrei
