Fix: Hash @indices can grow larger than Int32::MAX bytes (#15347)

`Hash` can theoretically hold up to `Int32::MAX` entries, but `@indices` grows faster. Using `Int32` maths to calculate the memory size for `@indices` limits capacity to `Int32::MAX // 4`. This patch uses `Int64` maths instead to enable bigger hash sizes.

Co-authored-by: Johannes Müller <[email protected]>

spec/manual/hash_large_spec.cr

@@ -0,0 +1,8 @@
+require "spec"
+
+it "creates Hash at maximum capacity" do
+  # we don't try to go as high as Int32::MAX because it would allocate 18GB of
+  # memory in total. This already tests for Int32 overflows while 'only' needing
+  # 4.5GB of memory.
+  Hash(Int32, Int32).new(initial_capacity: (Int32::MAX // 4) + 1)
+end

src/hash.cr

@@ -236,12 +236,14 @@ class Hash(K, V)
       # Translate initial capacity to the nearest power of 2, but keep it a minimum of 8.
       if initial_capacity < 8
         initial_entries_size = 8
+      elsif initial_capacity > 2**30
+        initial_entries_size = Int32::MAX
       else
         initial_entries_size = Math.pw2ceil(initial_capacity)
       end
 
       # Because we always keep indice_size >= entries_size * 2
-      initial_indices_size = initial_entries_size * 2
+      initial_indices_size = initial_entries_size.to_u64 * 2
 
       @entries = malloc_entries(initial_entries_size)
 
@@ -830,7 +832,7 @@ class Hash(K, V)
 
   # The actual number of bytes needed to allocate `@indices`.
   private def indices_malloc_size(size)
-    size * @indices_bytesize
+    size.to_u64 * @indices_bytesize
   end
 
   # Reallocates `size` number of indices for `@indices`.