Reputation: 936
How does Go calculate a hash for keys in a map? Is it truly unique and is it available for use in other structures?
I imagine it's easy for primitive keys like int
or immutable string
but it seems nontrivial for composite structures.
Upvotes: 18
Views: 8222
Reputation: 239652
Since Go 1.14, the go standard library provides the hash/maphash package. The hash functions in this package aren't guaranteed to be the same ones used by Go maps (but it appears that they are, which makes sense); they are guaranteed to be good functions for implementing hashmaps and the like.
hash/maphash only operates on strings or byte slices, so it's still up to you to figure out how to serialize a composite data structure into bytes for hashing purposes.
Upvotes: 7
Reputation: 239652
The language spec doesn't say, which means that it's free to change at any time, or differ between implementations.
The hash algorithm varies somewhat between types and platforms. As of now: On x86 (32 or 64 bit) if the CPU supports AES instructions, the runtime uses aeshash
, a hash built on AES primitives, otherwise it uses a function "inspired by" xxHash and cityhash, but different from either. There are different variants for 32-bit and 64-bit systems. Most types use a simple hash of their memory contents, but floating-point types have code to ensure that 0 and -0 hash equally (since they compare equally) and NaNs hash randomly (since two NaNs are never equal). Since complex types are built from floats, their hashes are composed from the hashes of their two floating-point parts. And an interface's hash is the hash of the value stored in the interface, and not the interface header itself.
Upvotes: 22