Issue #9188 has been updated by tmm1 (Aman Gupta).


@charliesome helped me with my compiler patch. It adds a new opt_aset_str instruction that handles the string literal hash_aset case (hsh["abc"]=).

Unfortunately, it did not make a big difference in long-lived strings generated in our app. The patch follows for posterity.

diff --git a/compile.c b/compile.c
index 812f692..1cc038c 100644
--- a/compile.c
+++ b/compile.c
@@ -5290,6 +5290,24 @@ iseq_compile_each(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE * node, int poped)
 	break;
       }
       case NODE_ATTRASGN:{
+	if (node->nd_mid == idASET &&
+	    node->nd_recv != (NODE *)1 &&
+	    node->nd_args &&
+	    nd_type(node->nd_args) == NODE_ARRAY &&
+	    node->nd_args->nd_alen == 2 &&
+	    nd_type(node->nd_args->nd_head) == NODE_STR)
+	{
+	    COMPILE(ret, "recv", node->nd_recv);
+	    COMPILE(ret, "value", node->nd_args->nd_next->nd_head);
+	    ADD_INSN2(ret, line, opt_aset_str,
+		      new_callinfo(iseq, idASET, 2, 0, 0),
+		      rb_fstring(node->nd_args->nd_head->nd_lit));
+	    if (poped) {
+		ADD_INSN(ret, line, pop);
+	    }
+	    break;
+	}
+
 	DECL_ANCHOR(recv);
 	DECL_ANCHOR(args);
 	VALUE flag = 0;
diff --git a/hash.c b/hash.c
index 1321b83..8bbd569 100644
--- a/hash.c
+++ b/hash.c
@@ -2368,7 +2368,7 @@ static VALUE rb_hash_compare_by_id_p(VALUE hash);
  *     h1["a"]        #=> 100
  *     h1.compare_by_identity
  *     h1.compare_by_identity? #=> true
- *     h1["a"]        #=> nil  # different objects.
+ *     h1["a".dup]    #=> nil  # different objects.
  *     h1[:c]         #=> "c"  # same symbols are all same.
  *
  */
diff --git a/insns.def b/insns.def
index 63a36b3..dcecfdc 100644
--- a/insns.def
+++ b/insns.def
@@ -1903,6 +1903,27 @@ opt_aset
 
 /**
   @c optimize
+  @e recv[str] = set
+  @j ???????????????? recv[str] = set???
+ */
+DEFINE_INSN
+opt_aset_str
+(CALL_INFO ci, VALUE key)
+(VALUE recv, VALUE val)
+(VALUE val)
+{
+    if (!SPECIAL_CONST_P(recv) && RBASIC_CLASS(recv) == rb_cHash && BASIC_OP_UNREDEFINED_P(BOP_ASET, HASH_REDEFINED_OP_FLAG)) {
+	rb_hash_aset(recv, key, val);
+    } else {
+	PUSH(recv);
+	PUSH(rb_str_resurrect(key));
+	PUSH(val);
+	CALL_SIMPLE_METHOD(recv);
+    }
+}
+
+/**
+  @c optimize
   @e optimized length
   @j ???????????????? recv.length()???
  */
diff --git a/test/ruby/test_hash.rb b/test/ruby/test_hash.rb
index 4244b86..ef606bf 100644
--- a/test/ruby/test_hash.rb
+++ b/test/ruby/test_hash.rb
@@ -209,6 +209,13 @@ class TestHash < Test::Unit::TestCase
     assert_equal(256,     h[z])
   end
 
+  def test_ASET_fstring_key
+    a, b = {}, {}
+    a["abc"] = 1
+    b["abc"] = 1
+    assert_same a.keys[0], b.keys[0]
+  end
+
   def test_NEWHASH_fstring_key
     a = {"ABC" => :t}
     b = {"ABC" => :t}
@@ -897,7 +904,7 @@ class TestHash < Test::Unit::TestCase
     h = @cls[]
     h.compare_by_identity
     h["a"] = 1
-    h["a"] = 2
+    h["a".dup] = 2
     assert_equal(["a",1], h.assoc("a"))
   end
 

----------------------------------------
misc #9188: r43870 make benchmark/bm_so_k_nucleotide.rb slow
https://bugs.ruby-lang.org/issues/9188#change-43505

Author: authorNari (Narihiro Nakamura)
Status: Closed
Priority: Normal
Assignee: tmm1 (Aman Gupta)
Category: core
Target version: current: 2.1.0


Hi.

I think r43870 make benchmark/bm_so_k_nucleotide.rb slow.

r43870
% time ./miniruby ./benchmark/bm_so_k_nucleotide.rb
./miniruby ./benchmark/bm_so_k_nucleotide.rb  1.70s user 0.01s system 99% cpu 1.718 total

r43869
% time ./miniruby ./benchmark/bm_so_k_nucleotide.rb
./miniruby ./benchmark/bm_so_k_nucleotide.rb  1.52s user 0.03s system 99% cpu 1.559 total


-- 
http://bugs.ruby-lang.org/