filedictrb 0.1.4 → 0.2.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: f023c0d5acca6fc0f6355bae7c34f61eb5682f9f4652f83dd8befefac6f2c050
-  data.tar.gz: de7ba4f1fcf57e21771312bf9bb8daacdfc39a044e9ef263a81494bb6e976dca
+  metadata.gz: 7cb1c1d801bf8943d6493a6c40be882c5d47c58166c56b6860abf8b98fc14a98
+  data.tar.gz: 394f0febd55f827bfb3aafb3559ba89b6b458b3a7ed2c3b6d857d900e38b7374
 SHA512:
-  metadata.gz: 2cb1f1f1a43a0494aa95f8d31d0b1b490f63ce07c57bcc1e6af2003064656784034064cebefd8247b886174e3e57c600c4464e2eba69a61cf7720c6cee9fcc1e
-  data.tar.gz: 3f4f4dbf014f9af21ad104a3282f09ce7d07c22f70c60e5e33b43c58159bb403567e42116896091062a527bd881524d15db9b8724a16065395b0cdd6cdb715c1
+  metadata.gz: 4a700778e49bd15c6686b9457b7a7987f4d6c8b47be4abf805a6a09d95156ce71ff8e15e7cb50ffcfa11f96e78f3eb633c19d10bf472c6f7963deb5ae658f9f4
+  data.tar.gz: dc233039e2c4d0e2aade78efc7cd6b55ab088cbdeaa373d9219137ae1241585dbe32b4e5984eeb402c52a32a027ee14f2950fa5540b3a3b932a34d13ff834b8c

data/ext/filedict/filedict.h

@@ -120,15 +120,17 @@ static void filedict_deinit(filedict_t *filedict) {
  * This computes the size of the entire filedict file given an initial bucket count and hashmap count.
  */
 static size_t filedict_file_size(size_t initial_bucket_count, size_t hashmap_count) {
-    size_t result = sizeof(filedict_header_t);
-    size_t i;
-
-    for (i = 0; i < hashmap_count; ++i) {
-        /* Bucket count is multiplied by 2 for each additional hashmap. */
-        result += (initial_bucket_count << i) * sizeof(filedict_bucket_t);
-    }
-
-    return result;
+    /*
+     * We used to size each additional hashmap at 2x the previous, but realistically it seems that
+     * most resizes are triggered by keys that are ridiculously large, not by mass collision.
+     *
+     * A more proper fix might be to re-structure the whole filedict. We could keep the existing
+     * hashmap structure, but with buckets that expand dynamically. This would require each bucket
+     * to contain a "pointer" to the next bucket object if present.
+     *
+     * For now, it's easiser to just keep the hashmap duplication without the size doubling.
+     */
+    return sizeof(filedict_header_t) + initial_bucket_count * hashmap_count * sizeof(filedict_bucket_t);
 }
 
 /*
@@ -289,7 +291,6 @@ try_again:
 
         ++hashmap_i;
         hashmap += bucket_count;
-        bucket_count = (bucket_count << 1);
     }
 
     /*
@@ -409,7 +410,7 @@ static int filedict_read_advance_hashmap(filedict_read_t *read) {
 
     filedict_bucket_t *hashmap = filedict->data + offset;
 
-    read->bucket_count = (size_t)header->initial_bucket_count << read->hashmap_i;
+    read->bucket_count = (size_t)header->initial_bucket_count;
     read->bucket = &hashmap[read->key_hash % read->bucket_count];
     read->entry = &read->bucket->entries[0];
 

data/lib/filedict/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Filedict
-  VERSION = "0.1.4"
+  VERSION = "0.2.0"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: filedictrb
 version: !ruby/object:Gem::Version
-  version: 0.1.4
+  version: 0.2.0
 platform: ruby
 authors:
 - Nigel Baillie
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2022-04-04 00:00:00.000000000 Z
+date: 2022-04-06 00:00:00.000000000 Z
 dependencies: []
 description:
 email:
@@ -58,7 +58,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.1.6
+rubygems_version: 3.3.3
 signing_key:
 specification_version: 4
 summary: Uses filedict to emulate a file-backed Hash<Set<String>>