On 03 November 2005 17:08, Jan-Willem Maessen wrote:
I've recently been experimenting with unsafeFreeze/unsafeThaw in GHC. Judicious use of these functions vastly reduces GC overhead in Data.HashTable. However, a slightly mis-timed GC will cause the whole mess to crash. I am attempting to understand the invariants required to safely use unsafeFreeze/unsafeThaw.
I believe the following usage ought to be 100% safe:
1) Take the last and only reference to a mutable array, and call unsafeFreeze to obtain an immutable array.
2) Take the last and only reference to an immutable array, and call unsafeThaw to obtain a mutable array.
I would like it to be the case that no usage of unsafe{Freeze,Thaw} can crash the GC. That perhaps isn't true right now, but I'd like to understand why not. Multiple unsafeThaws on the same array might cause problems in 6.4.1 (but not the HEAD), due to the way the old generation mutable list works. That's fixable, if it is the problem. Multiple unsafeFreezes should be harmless. Can you give me a test program? Cheers, Simon