From ed8b452e37322e66eb94dafa36f2b5da8f5ca44a Mon Sep 17 00:00:00 2001
From: nobu <nobu@b2dd03c8-39d4-4d8f-98ff-823fe69b080e>
Date: Tue, 29 Sep 2015 07:37:40 +0000
Subject: [PATCH] compile.c: fix performance of strconcat

* compile.c (compile_dstr_fragments): fix performance by omitting
  the first empty string only for keeping literal encoding if
  other literals are too.  [ruby-core:70930] [Bug #11556]
* string.c (rb_str_append_literal): append but keep encoding non
  US-ASCII.

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@51970 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
---
 ChangeLog |  9 +++++++++
 compile.c | 13 +++++++++++++
 insns.def |  2 +-
 string.c  | 12 ++++++++++++
 vm.c      |  2 ++
 5 files changed, 37 insertions(+), 1 deletion(-)

diff --git a/ChangeLog b/ChangeLog
index 47bc21039b..be8d0d1aab 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+Tue Sep 29 16:37:29 2015  Nobuyoshi Nakada  <nobu@ruby-lang.org>
+
+	* compile.c (compile_dstr_fragments): fix performance by omitting
+	  the first empty string only for keeping literal encoding if
+	  other literals are too.  [ruby-core:70930] [Bug #11556]
+
+	* string.c (rb_str_append_literal): append but keep encoding non
+	  US-ASCII.
+
 Mon Sep 28 17:40:17 2015  Shugo Maeda  <shugo@ruby-lang.org>
 
 	* lib/net/ftp.rb (mtime): use usec instead of fractions to parse
diff --git a/compile.c b/compile.c
index 9928aaa44c..9c59b595d7 100644
--- a/compile.c
+++ b/compile.c
@@ -787,6 +787,12 @@ FIRST_ELEMENT(LINK_ANCHOR *anchor)
     return anchor->anchor.next;
 }
 
+static LINK_ELEMENT *
+LAST_ELEMENT(LINK_ANCHOR *anchor)
+{
+    return anchor->last;
+}
+
 static LINK_ELEMENT *
 POP_ELEMENT(ISEQ_ARG_DECLARE LINK_ANCHOR *anchor)
 {
@@ -2329,6 +2335,7 @@ compile_dstr_fragments(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE *node, int *cntp)
 {
     NODE *list = node->nd_next;
     VALUE lit = node->nd_lit;
+    LINK_ELEMENT *first_lit = 0;
     int cnt = 0;
 
     debugp_param("nd_lit", lit);
@@ -2337,6 +2344,7 @@ compile_dstr_fragments(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE *node, int *cntp)
 	if (RB_TYPE_P(lit, T_STRING))
 	    lit = node->nd_lit = rb_fstring(node->nd_lit);
 	ADD_INSN1(ret, nd_line(node), putobject, lit);
+	if (RSTRING_LEN(lit) == 0) first_lit = LAST_ELEMENT(ret);
     }
 
     while (list) {
@@ -2344,6 +2352,7 @@ compile_dstr_fragments(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE *node, int *cntp)
 	if (nd_type(node) == NODE_STR) {
 	    node->nd_lit = rb_fstring(node->nd_lit);
 	    ADD_INSN1(ret, nd_line(node), putobject, node->nd_lit);
+	    lit = Qnil;
 	}
 	else {
 	    COMPILE(ret, "each string", node);
@@ -2351,6 +2360,10 @@ compile_dstr_fragments(rb_iseq_t *iseq, LINK_ANCHOR *ret, NODE *node, int *cntp)
 	cnt++;
 	list = list->nd_next;
     }
+    if (NIL_P(lit) && first_lit) {
+	REMOVE_ELEM(first_lit);
+	--cnt;
+    }
     *cntp = cnt;
 
     return COMPILE_OK;
diff --git a/insns.def b/insns.def
index 7eb322120b..a923d3a4d6 100644
--- a/insns.def
+++ b/insns.def
@@ -370,7 +370,7 @@ concatstrings
     val = rb_str_resurrect(TOPN(i));
     while (i-- > 0) {
 	const VALUE v = TOPN(i);
-	rb_str_append(val, v);
+	rb_str_append_literal(val, v);
     }
     POPN(num);
 }
diff --git a/string.c b/string.c
index e93af5cece..728676c159 100644
--- a/string.c
+++ b/string.c
@@ -2460,6 +2460,18 @@ rb_str_append(VALUE str, VALUE str2)
     return rb_str_buf_append(str, str2);
 }
 
+VALUE
+rb_str_append_literal(VALUE str, VALUE str2)
+{
+    int encidx = rb_enc_get_index(str2);
+    rb_str_buf_append(str, str2);
+    if (encidx != ENCINDEX_US_ASCII) {
+	if (rb_enc_get_index(str) == ENCINDEX_US_ASCII)
+	    rb_enc_associate_index(str, encidx);
+    }
+    return str;
+}
+
 /*
  *  call-seq:
  *     str << integer       -> str
diff --git a/vm.c b/vm.c
index 92b33f3ffe..ee9f4076c1 100644
--- a/vm.c
+++ b/vm.c
@@ -21,6 +21,8 @@
 #include "probes.h"
 #include "probes_helper.h"
 
+VALUE rb_str_append_literal(VALUE str, VALUE str2);
+
 static inline VALUE *
 VM_EP_LEP(VALUE *ep)
 {