On Tue, Feb 28, 2012 at 12:04:45PM -0800, Walter Bright wrote: > On 2/28/2012 7:20 AM, H. S. Teoh wrote: > >That's strange, because I'm clearly seeing a difference in hash > >value. Calling map1's keyti.getHash(key) yields a different value > >from map2's keyti.getHash(key). > > The thing to do then is look at those two getHash functions, and see > how they are computing it.
Question: how does the compiler generate TypeInfo objects for const(int)[] and immutable(int)[]? I think that's where the bug lies. I wrote a program to manually call rt.util.hash.hashOf by casting const and immutable int arrays' ptr into void* and passing in the data directly (this is what typeid(int[]).getHash() does), and the hash values are consistent with int[]. However, calling typeid(const(int)[]).getHash and typeid(immutable(int)[]).getHash produces a different value (the same between const and immutable, but different from unqualified int[]). So it looks like const and immutable int[] are using some kind of generic array typeinfo code? But I have trouble finding where in the druntime code this is. In any case, the generic array typeinfo code is obviously doing something different from the hand-coded typeid(int[]), that's why the hash value is inconsistent. T -- Маленькие детки - маленькие бедки.
