Update hash mixing behavior - address #889

Try to have better behavior when mixing sub-hashes that are not uniform and
randomly distributed. Premultiply by a large prime before mixing to
"spread entropy" if it is concentrated in a certain subset of bits.

.github/workflows/release.yml

@@ -50,5 +50,6 @@ jobs:
         with:
           draft: true
           files: |
+            ./dist/*.zip
             ./*.zip
             ./*.msi

CHANGELOG.md

@@ -1,8 +1,11 @@
 # Changelog
 All notable changes to this project will be documented in this file.
 
+## Unreleased - ???
+- Update hash function.
+
 ## 1.19.1 - 2021-12-04
-- Add an optional `prefix` paramater to `debug/stacktrace` to allow printing prettier error messages.
+- Add an optional `prefix` parameter to `debug/stacktrace` to allow printing prettier error messages.
 - Remove appveyor for CI pipeline
 - Fixed a bug that prevented sending threaded abstracts over threaded channels.
 - Fix bug in the `map` function with arity at least 3.

meson.build

@@ -20,7 +20,7 @@
 
 project('janet', 'c',
   default_options : ['c_std=c99', 'build.c_std=c99', 'b_lundef=false', 'default_library=both'],
-  version : '1.19.1')
+  version : '1.19.2')
 
 # Global settings
 janet_path = join_paths(get_option('prefix'), get_option('libdir'), 'janet')

src/conf/janetconf.h

@@ -5,9 +5,9 @@
 
 #define JANET_VERSION_MAJOR 1
 #define JANET_VERSION_MINOR 19
-#define JANET_VERSION_PATCH 1
+#define JANET_VERSION_PATCH 2
-#define JANET_VERSION_EXTRA ""
+#define JANET_VERSION_EXTRA "-dev"
-#define JANET_VERSION "1.19.1"
+#define JANET_VERSION "1.19.2-dev"
 
 /* #define JANET_BUILD "local" */
 

src/core/util.c

@@ -224,13 +224,16 @@ int32_t janet_string_calchash(const uint8_t *str, int32_t len) {
 
 #endif
 
+uint32_t janet_hash_mix(uint32_t input, uint32_t more) {
+    return input ^ ((more * 2119589369u) + 0x9e3779b9 + (input << 6) + (input >> 2));
+}
+
 /* Computes hash of an array of values */
 int32_t janet_array_calchash(const Janet *array, int32_t len) {
     const Janet *end = array + len;
-    uint32_t hash = 0;
+    uint32_t hash = 33;
     while (array < end) {
-        uint32_t elem = janet_hash(*array++);
+        hash = janet_hash_mix(hash, janet_hash(*array++));
-        hash ^= elem + 0x9e3779b9 + (hash << 6) + (hash >> 2);
     }
     return (int32_t) hash;
 }
@@ -238,10 +241,10 @@ int32_t janet_array_calchash(const Janet *array, int32_t len) {
 /* Computes hash of an array of values */
 int32_t janet_kv_calchash(const JanetKV *kvs, int32_t len) {
     const JanetKV *end = kvs + len;
-    uint32_t hash = 0;
+    uint32_t hash = 33;
     while (kvs < end) {
-        hash ^= janet_hash(kvs->key) + 0x9e3779b9 + (hash << 6) + (hash >> 2);
+        hash = janet_hash_mix(hash, janet_hash(kvs->key));
-        hash ^= janet_hash(kvs->value) + 0x9e3779b9 + (hash << 6) + (hash >> 2);
+        hash = janet_hash_mix(hash, janet_hash(kvs->value));
         kvs++;
     }
     return (int32_t) hash;

src/core/util.h

@@ -56,6 +56,7 @@
 } while (0)
 
 /* Utils */
+uint32_t janet_hash_mix(uint32_t input, uint32_t more);
 #define janet_maphash(cap, hash) ((uint32_t)(hash) & (cap - 1))
 int janet_valid_utf8(const uint8_t *str, int32_t len);
 int janet_is_symbol_char(uint8_t c);