From 6d17a3e6472b4c6ce81f5308fc07050855114410 Mon Sep 17 00:00:00 2001
From: Yusuke Endoh <mame@ruby-lang.org>
Date: Wed, 16 Jul 2025 17:20:21 +0900
Subject: [PATCH] Prevent ObjectSpace.count_objects from allocating extra
 arrays

`ObjectSpace.count_objects` could cause an unintended array allocation.
It returns a hash like `{ :T_ARRAY => 100, :T_STRING => 100, ... }`, so
it creates the key symbol (e.g., `:T_STRING`) for the first time. On
rare occations, this symbol creation internally allocates a new array
for symbol management.

This led to a problematic side effect where calling `count_objects`
twice in a row could produce inconsistent results: the first call would
trigger the hidden array allocation, and the second call would then
report an increased count for `:T_ARRAY`.

This behavior caused test failures in `test/ruby/test_allocation.rb`,
which performs a baseline measurement before an operation and then
asserts the exact number of new allocations.

https://rubyci.s3.amazonaws.com/openbsd-current/ruby-master/log/20250716T053005Z.fail.html.gz

>   1) Failure:
> TestAllocation::ProcCall::WithBlock#test_ruby2_keywords [...]:
> Expected 1 array allocations for "r2k.(1, a: 2, &block)", but 2 arrays allocated.

This change resolves the issue by pre-interning all key symbols used by
`ObjectSpace.count_objects` before its counting. This eliminates the
side effect and ensures the stability of allocation-sensitive tests.

Co-authored-by: Koichi Sasada <ko1@atdot.net>
---
 gc.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/gc.c b/gc.c
index 4a3623cd75..10bce22639 100644
--- a/gc.c
+++ b/gc.c
@@ -2498,6 +2498,7 @@ count_objects(int argc, VALUE *argv, VALUE os)
 {
     struct count_objects_data data = { 0 };
     VALUE hash = Qnil;
+    VALUE types[T_MASK + 1];
 
     if (rb_check_arity(argc, 0, 1) == 1) {
         hash = argv[0];
@@ -2505,6 +2506,10 @@ count_objects(int argc, VALUE *argv, VALUE os)
             rb_raise(rb_eTypeError, "non-hash given");
     }
 
+    for (size_t i = 0; i <= T_MASK; i++) {
+        types[i] = type_sym(i);
+    }
+
     rb_gc_impl_each_object(rb_gc_get_objspace(), count_objects_i, &data);
 
     if (NIL_P(hash)) {
@@ -2517,9 +2522,9 @@ count_objects(int argc, VALUE *argv, VALUE os)
     rb_hash_aset(hash, ID2SYM(rb_intern("FREE")), SIZET2NUM(data.freed));
 
     for (size_t i = 0; i <= T_MASK; i++) {
-        VALUE type = type_sym(i);
-        if (data.counts[i])
-            rb_hash_aset(hash, type, SIZET2NUM(data.counts[i]));
+        if (data.counts[i]) {
+            rb_hash_aset(hash, types[i], SIZET2NUM(data.counts[i]));
+        }
     }
 
     return hash;