[ruby-trunk - Feature #9816] 文字列内の数字を数値として比較するメソッド

Issue #9816 has been updated by Yui NARUSE.

熟考の結果、Gem::Versionと仕様をあわせました。
理由は、

  • Gem::Versionでこれを使ってくれればオブジェクトの生成数が減る
  • 2.2.0-preview1のようなRubyのバージョンの比較ができる
    からです。

順序のイメージは Prereleases sort between real releases (newest to oldest)
のような感じです

  1. 1.0
  2. 1.0.b1
  3. 1.0.a.2
  4. 0.9
diff --git a/string.c b/string.c
index bec0bfd..e2b3c6f 100644
--- a/string.c
+++ b/string.c
@@ -2605,6 +2605,232 @@ rb_str_casecmp(VALUE str1, VALUE str2)
     return INT2FIX(-1);
 }

+static int
+version_string_p(VALUE str)
+{
+    const char *p = RSTRING_PTR(str);
+    const char *e = RSTRING_END(str);
+
+    if (!rb_enc_asciicompat(STR_ENC_GET(str))) return FALSE;
+
+    if (!ISDIGIT(*p)) return FALSE;
+    do { if (++p >= e) return TRUE; } while (ISDIGIT(*p));
+
+    while (*p == '.') {
+  if (++p >= e) return FALSE;
+  if (!ISALNUM(*p)) return FALSE;
+  do { if (++p >= e) return TRUE; } while (ISALNUM(*p));
+    }
+
+    if (*p != '-') return FALSE;
+     do {
+  if (++p >= e) return FALSE;
+  if (!ISALNUM(*p) && *p != '-') return FALSE;
+  do { if (++p >= e) return TRUE; } while (ISALNUM(*p) || *p == '-');
+    } while (*p == '.');
+
+    return FALSE;
+}
+
+/* return value: whether end of nueric part is EOS
+ * sp: first nonzero digit
+ * ep: end of digits
+ */
+static void
+search_numerical_str(const char **sp, const char **ep)
+{
+    const char *p = *sp;
+    const char *e = *ep;
+    assert(p < e);
+    for (;;) {
+  if (*p != '0') break;
+  p++;
+  if (p == e) {
+      *sp = p;
+      goto finish;
+  }
+    }
+    *sp = p;
+    assert(p < e);
+    for (;;) {
+  if (!ISDIGIT(*p)) break;
+  p++;
+  if (p == e) {
+      goto finish;
+  }
+    }
+finish:
+    *ep = p;
+    return;
+}
+
+static VALUE
+numerical_compare(const char **pp1, const char *p1end, const char 
**pp2, const char *p2end)
+{
+    const char *s1 = *pp1, *p1=p1end, *s2 = *pp2, *p2=p2end;
+    ptrdiff_t len1, len2;
+    int r;
+
+    search_numerical_str(&s1, &p1);
+    search_numerical_str(&s2, &p2);
+
+    /* compre digits length */
+    len1 = p1 - s1;
+    len2 = p2 - s2;
+    if (len1 != len2) return INT2FIX(len1 < len2 ? -1 : 1);
+
+    /* compre numeric value */
+    r = memcmp(s1, s2, len1);
+    if (r) return r < 0 ? INT2FIX(-1) : INT2FIX(1);
+
+    *pp1 = p1;
+    *pp2 = p2;
+    return Qnil;
+}
+
+/*
+ *  call-seq:
+ *     str.versioncmp(other_str)   -> -1, 0, +1 or nil
+ *
+ *  Compare strings as version strings.
+ *
+ *     "a1".versioncmp("a1")            #=> 0
+ *     "aa".versioncmp("a1")            #=> 1
+ *     "a1".versioncmp("aa")            #=> -1
+ *     "a1".versioncmp("a01")           #=> -1
+ *     "2.1.2".numericcmp("2.1.10")     #=> 1
+ */
+
+static VALUE
+rb_str_versioncmp(VALUE str1, VALUE str2)
+{
+    const char *p, *pe, *q, *qe;
+
+    StringValue(str2);
+    if (!version_string_p(str1)) {
+  rb_raise(rb_eArgError, "receiver is not version string 
'%+"PRIsVALUE"'", str1);
+    }
+    if (!version_string_p(str2)) {
+  rb_raise(rb_eArgError, "argument is not version string 
'%+"PRIsVALUE"'", str2);
+    }
+
+    p = RSTRING_PTR(str1); pe = RSTRING_END(str1);
+    q = RSTRING_PTR(str2); qe = RSTRING_END(str2);
+
+    for (;;) {
+  if (*p == '-') {
+hyphen_left:
+      if (*q == '-') goto next_char;
+      while (*q == '.') {
+    if (++q == qe) return INT2FIX(1);
+      }
+      if (*q != 'p') return INT2FIX(ISDIGIT(*q) || 'p' < *q ? -1 : 1);
+      if (++q == qe) return INT2FIX(1);
+      if (*q != 'r') return INT2FIX(ISDIGIT(*q) || 'r' < *q ? -1 : 1);
+      if (++q == qe) return INT2FIX(1);
+      if (*q != 'e') return INT2FIX(ISDIGIT(*q) || 'e' < *q ? -1 : 1);
+      if (++q == qe) return INT2FIX(1);
+      if (*q != '.') {
+    if (*q == '-') {
+        p++;
+        goto hyphen_right;
+    }
+    else if (ISALPHA(*q)) return INT2FIX(-1);
+    q--; /* DIGIT */
+      }
+  }
+  else if (*q == '-') {
+hyphen_right:
+      if (*p == '-') goto next_char;
+      while (*p == '.') {
+    if (++p == pe) return INT2FIX(-1);
+      }
+      if (*p != 'p') return INT2FIX(ISDIGIT(*p) || 'p' < *p ? 1 : -1);
+      if (++p == pe) return INT2FIX(-1);
+      if (*p != 'r') return INT2FIX(ISDIGIT(*p) || 'r' < *p ? 1 : -1);
+      if (++p == pe) return INT2FIX(-1);
+      if (*p != 'e') return INT2FIX(ISDIGIT(*p) || 'e' < *p ? 1 : -1);
+      if (++p == pe) return INT2FIX(-1);
+      if (*p == '-') {
+    q++;
+    goto hyphen_left;
+      }
+      else if (ISALPHA(*p)) return INT2FIX(1);
+      else if (ISDIGIT(*p)) {
+    p--; /* DIGIT */
+      }
+  }
+  else if (ISDIGIT(*p)) {
+      if (ISDIGIT(*q)) {
+    VALUE r = numerical_compare(&p, pe, &q, qe);
+    if(!NIL_P(r)) return r;
+    goto incremented;
+      }
+      else {
+    return INT2FIX(1);
+      }
+  }
+  else if (ISDIGIT(*q)) {
+      return INT2FIX(-1);
+  }
+  else if (ISALPHA(*p)) {
+      if (ISALPHA(*q)) {
+    for (;;) {
+        if (*p != *q) return INT2FIX(*p < *q ? -1 : 1);
+        p++;
+        q++;
+        if (p == pe) {
+      if (q == qe) return INT2FIX(0);
+      if (ISALPHA(*q)) return INT2FIX(-1);
+      goto incremented;
+        }
+        else if (q == qe) {
+      if (ISALPHA(*p)) return INT2FIX(1);
+      goto incremented;
+        }
+        else if (ISALPHA(*p)) {
+      if (!ISALPHA(*q)) return INT2FIX(1);
+        }
+        else if (ISALPHA(*q)) return INT2FIX(-1);
+        else goto incremented;
+    }
+    continue;
+      }
+      else return INT2FIX(1);
+  }
+  else if (ISALPHA(*q)) {
+      return INT2FIX(-1);
+  }
+  else rb_bug("%s %s",p,q);
+
+next_char:
+  p++;
+  q++;
+
+incremented:
+  while (*p == '.' && ++p != pe);
+  while (*q == '.' && ++q != qe);
+  if (p == pe) {
+      if (q == qe) return INT2FIX(0);
+      if (ISDIGIT(*q)) {
+    return INT2FIX(-1);
+      }
+      else /*if (ISALPHA(*q) || *q == '-')*/ {
+    return INT2FIX(1);
+      }
+  }
+  else if (q == qe) {
+      if (ISDIGIT(*p)) {
+    return INT2FIX(1);
+      }
+      else /*if (ISALPHA(*p) || *p == '-')*/ {
+    return INT2FIX(-1);
+      }
+  }
+    }
+    UNREACHABLE;
+}
+
 #define rb_str_index(str, sub, offset) rb_strseq_index(str, sub, 
offset, 0)

 static long
@@ -8778,6 +9004,7 @@ Init_String(void)
     rb_define_method(rb_cString, "eql?", rb_str_eql, 1);
     rb_define_method(rb_cString, "hash", rb_str_hash_m, 0);
     rb_define_method(rb_cString, "casecmp", rb_str_casecmp, 1);
+    rb_define_method(rb_cString, "versioncmp", rb_str_versioncmp, 1);
     rb_define_method(rb_cString, "+", rb_str_plus, 1);
     rb_define_method(rb_cString, "*", rb_str_times, 1);
     rb_define_method(rb_cString, "%", rb_str_format_m, 1);
diff --git a/test/ruby/test_string.rb b/test/ruby/test_string.rb
index e8decc0..9e92fb7 100644
--- a/test/ruby/test_string.rb
+++ b/test/ruby/test_string.rb
@@ -2112,6 +2112,41 @@ def test_casecmp
     assert_equal(1, "\u3042B".casecmp("\u3042a"))
   end

+  def test_versioncmp
+    require "rubygems"
+    ary = %w[
+      1
+      2
+      10
+      1.a
+      1.a-a
+      1.a--
+      1.a--.-
+      1.a-1
+      1.a.q
+      1.a--a
+      1.a--1
+      1.a.pre.a
+      1.a-pre.a
+      1.a.pre-a
+      1.a1
+      1.a2
+      1.aa
+      1.b
+      1.01
+      1.1
+      1.1a
+      1.1-a
+      1.1-b
+      1.1q
+      1.2
+      1.10
+    ]
+    ary.product(ary) do |a, b|
+      assert_equal(Gem::Version.new(a)<=>Gem::Version.new(b), 
a.versioncmp(b), "#{a.dump}, #{b.dump}")
+    end
+  end
+
   def test_upcase2
     assert_equal("\u3042AB", "\u3042aB".upcase)
   end

Feature #9816: 文字列内の数字を数値として比較するメソッド

  • Author: Yui NARUSE
  • Status: Assigned
  • Priority: Normal
  • Assignee: Yukihiro M.
  • Category: core
  • Target version:

文字列内の数字を数値として比較するメソッドを追加しませんか

そのような比較は一般的な用途としてはGUIシェルのファイラーが比較に用いており、
Windows では StrCmpLogicalW が、OS X では
NSString:compare:options:へのNSNumericSearch定数が提供されています。

上記のような処理自体はさほど難しいものではありませんが、Rubyレベルで実装すると大量のオブジェクトを作ってしまいます。
例えば
Gem::Version.new("2.1.10".freeze)<=>Gem::Version.new("2.1.9".freeze)
は47個、
"2.1.10".freeze.split('.').map(&:to_i)<=>"2.1.9".freeze.split('.').map(&:to_i)
だと16個のオブジェクトを作ります。
"2.1.10".freeze.numericcmp"2.1.9".freeze ならば、もちろんオブジェクトは一つも作りません。

なお、上記の例でも示唆していますが、本メソッドは Ruby のバージョン表記の TEENY が2桁になった場合の比較に用いることができます。

パッチは以下の通りです。
なお、メソッド名は String#numericcmp としています。
(String#casecmpを念頭に置いた)

diff --git a/string.c b/string.c
index c589c80..66f667f 100644
--- a/string.c
+++ b/string.c
@@ -2569,6 +2569,131 @@ rb_str_casecmp(VALUE str1, VALUE str2)
     return INT2FIX(-1);
 }

+VALUE
+numerical_compare(const char **pp1, const char *p1end, const char 
**pp2, const char *p2end)
+{
+    const char *s1 = *pp1, *p1, *s2 = *pp2, *p2;
+    ptrdiff_t len1, len2;
+    int r;
+
+    while (s1 < p1end && *s1 == '0') s1++;
+    p1 = s1;
+    while (p1 < p1end && ISDIGIT(*p1)) p1++;
+    len1 = p1 - s1;
+
+    while (s2 < p2end && *s2 == '0') s2++;
+    p2 = s2;
+    while (p2 < p2end && ISDIGIT(*p2)) p2++;
+    len2 = p2 - s2;
+
+    if (len1 != len2) {
+  return INT2FIX(len1 < len2 ? -1 : 1);
+    }
+
+    r = memcmp(s1, s2, len1);
+    if (r) return r < 0 ? INT2FIX(-1) : INT2FIX(1);
+
+    len1 = s1 - *pp1;
+    len2 = s2 - *pp2;
+    if (len1 != len2) {
+  return INT2FIX(len1 < len2 ? -1 : 1);
+    }
+
+    *pp1 = p1;
+    *pp2 = p2;
+    return Qnil;
+}
+
+/*
+ *  call-seq:
+ *     str.numericcmp(other_str)   -> -1, 0, +1 or nil
+ *
+ *  Variant of <code>String#<=></code>, which considers digits in 
strings
+ *  are numeric value..
+ *
+ *     "a1".numericcmp("a1")            #=> 0
+ *     "aa".numericcmp("a1")            #=> 1
+ *     "a1".numericcmp("aa")            #=> -1
+ *     "a1".numericcmp("a01")           #=> -1
+ *     "2.1.2".numericcmp("2.1.10")     #=> 1
+ */
+
+static VALUE
+rb_str_numericcmp(VALUE str1, VALUE str2)
+{
+    long len;
+    rb_encoding *enc;
+    const char *p1, *p1end, *p2, *p2end;
+
+    StringValue(str2);
+    enc = rb_enc_compatible(str1, str2);
+    if (!enc) {
+  return Qnil;
+    }
+
+    p1 = RSTRING_PTR(str1); p1end = RSTRING_END(str1);
+    p2 = RSTRING_PTR(str2); p2end = RSTRING_END(str2);
+    if (single_byte_optimizable(str1) && single_byte_optimizable(str2)) 
{
+  while (p1 < p1end && p2 < p2end) {
+      if (ISDIGIT(*p1)) {
+    if (ISDIGIT(*p2)) {
+        VALUE r = numerical_compare(&p1, p1end, &p2, p2end);
+        if (!NIL_P(r)) return r;
+    }
+    else {
+        return INT2FIX(-1);
+    }
+      }
+      else if (ISDIGIT(*p2)) {
+    return INT2FIX(1);
+      }
+      if (*p1 != *p2) return INT2FIX(*p1 < *p2 ? -1 : 1);
+      p1++;
+      p2++;
+  }
+    }
+    else {
+  while (p1 < p1end && p2 < p2end) {
+            int l1, c1 = rb_enc_ascget(p1, p1end, &l1, enc);
+            int l2, c2 = rb_enc_ascget(p2, p2end, &l2, enc);
+
+            if (0 <= c1 && 0 <= c2) {
+    if (ISDIGIT(*p1)) {
+        if (ISDIGIT(*p2)) {
+      VALUE r = numerical_compare(&p1, p1end, &p2, p2end);
+      if (!NIL_P(r)) return r;
+        }
+        else {
+      return INT2FIX(-1);
+        }
+    }
+    else if (ISDIGIT(*p2)) {
+        return INT2FIX(1);
+    }
+    if (*p1 != *p2) return INT2FIX(*p1 < *p2 ? -1 : 1);
+    p1++;
+    p2++;
+            }
+            else {
+                int r;
+                l1 = rb_enc_mbclen(p1, p1end, enc);
+                l2 = rb_enc_mbclen(p2, p2end, enc);
+                len = l1 < l2 ? l1 : l2;
+                r = memcmp(p1, p2, len);
+                if (r != 0)
+                    return INT2FIX(r < 0 ? -1 : 1);
+                if (l1 != l2)
+                    return INT2FIX(l1 < l2 ? -1 : 1);
+            }
+      p1 += l1;
+      p2 += l2;
+  }
+    }
+    if (RSTRING_LEN(str1) == RSTRING_LEN(str2)) return INT2FIX(0);
+    if (RSTRING_LEN(str1) > RSTRING_LEN(str2)) return INT2FIX(1);
+    return INT2FIX(-1);
+}
+
 static long
 rb_str_index(VALUE str, VALUE sub, long offset)
 {
@@ -8721,6 +8846,7 @@ Init_String(void)
     rb_define_method(rb_cString, "eql?", rb_str_eql, 1);
     rb_define_method(rb_cString, "hash", rb_str_hash_m, 0);
     rb_define_method(rb_cString, "casecmp", rb_str_casecmp, 1);
+    rb_define_method(rb_cString, "numericcmp", rb_str_numericcmp, 1);
     rb_define_method(rb_cString, "+", rb_str_plus, 1);
     rb_define_method(rb_cString, "*", rb_str_times, 1);
     rb_define_method(rb_cString, "%", rb_str_format_m, 1);
diff --git a/test/ruby/test_string.rb b/test/ruby/test_string.rb
index 8366424..f9c788b 100644
--- a/test/ruby/test_string.rb
+++ b/test/ruby/test_string.rb
@@ -2104,6 +2104,29 @@ class TestString < Test::Unit::TestCase
     assert_equal(1, "\u3042B".casecmp("\u3042a"))
   end

+  def test_numericcmp
+    assert_equal(-1, "2.1.0".numericcmp("2.1.1"))
+    assert_equal(-1, "2.1.9".numericcmp("2.1.10"))
+    assert_equal( 0, "a1".numericcmp("a1"))
+    assert_equal( 1, "aa".numericcmp("a1"))
+    assert_equal(-1, "a1".numericcmp("aa"))
+    assert_equal(-1, "a1".numericcmp("a01"))
+    assert_equal(-1, "a0001".numericcmp("a00001"))
+    assert_equal( 0, "a1a".numericcmp("a1a"))
+    assert_equal( 1, "a1b".numericcmp("a1a"))
+    assert_equal(-1, "a9a".numericcmp("a10a"))
+    assert_equal( 1, "b".numericcmp("a"))
+    assert_equal( 0, "\u30421".numericcmp("\u30421"))
+    assert_equal( 1, "\u3042\u3042".numericcmp("\u30421"))
+    assert_equal(-1, "\u30421".numericcmp("\u3042\u3042"))
+    assert_equal(-1, "\u30421".numericcmp("\u304201"))
+    assert_equal(-1, "\u30420001".numericcmp("\u304200001"))
+    assert_equal( 0, "\u30421\u3042".numericcmp("\u30421\u3042"))
+    assert_equal( 1, "\u30421\u3044".numericcmp("\u30421\u3042"))
+    assert_equal(-1, "\u30429\u3042".numericcmp("\u304210\u3042"))
+    assert_equal( 1, "\u3044".numericcmp("\u3042"))
+  end
+
   def test_upcase2
     assert_equal("\u3042AB", "\u3042aB".upcase)
   end