Hi, I have two questions about the reasons behind the design of Clojure's software transactional memory (STM):
(1) Why is snapshot isolation the default rather than linearizability? For those who don't know, the difference between the two guarantees is that snapshot isolation requires all reads to occur at a read point and all writes to occur at a later write point, whereas linearizability additionally requires those two points to be one and the same. Most of closure's design seems to prefer correctness over efficiency, but this decision places efficiency over correctness. Why not make the safe behavior of the current "ensure" the default with the one-character name and leave the more efficient but harder to understand "deref" behavior as something available when optimizing hot code? (2) Was the _principled_ form of open nesting in the paper titled "Open Nesting in Software Transactional Memory" ( http://www.cse.msu.edu/~cse914/Readings/openNestingInSoftwareTransMemory-ppopp07.pdf) considered for clojure's STM? That paper (which I have nothing to do with) deals with the issue of an STM sometimes thinking two operations conflict due to conflicting reads and writes even though from the perspective of abstract data types the transactions shouldn't conflict, For example if one transaction modifies key X in a hashmap and another modifies key Y those transactions shouldn't conflict unless those two hashmap operations are executing at the same time and don't execute correctly. If the two operations occur at different times then those two operations needn't conflict, even if they happen to both write the same data e.g. incrementing a counter of the number of keys in the table. This sort of principled open nesting solves the problem that clojure's "commute" does but also handles cases when an operation commutes with _some_ other operations but not _all_, e.g. two hashmap writes to different keys don't conflict with each other but writes to the same key do. Thanks in advance. Warren -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
