fast_xs 0.5 → 0.6

data/Rakefile

@@ -13,7 +13,7 @@ begin
   rev = Time.at(rev.split("\n")[1].to_i).strftime('%Y%m%d.%H%M%S')
 rescue
 end
-version ||= ENV['VERSION'] || '0.5' + (rev && rev.length > 0 ? ".#{rev}" : '')
+version ||= ENV['VERSION'] || '0.6' + (rev && rev.length > 0 ? ".#{rev}" : '')
 pkg = "#{name}-#{version}"
 bin = "*.{so,o}"
 archlib = "lib/#{::Config::CONFIG['arch']}"
@@ -45,7 +45,11 @@ spec = Gem::Specification.new do |s|
 end
 
 if ENV['VERSION']
+  # FIXME - look for hoe replacement...
+  prev = `git for-each-ref --format='%(refname)' refs/tags | tail -n1`.chomp
+  sh("git shortlog #{prev}..HEAD > History.txt");
   sh('git ls-files > Manifest.txt')
+  File.exists?('README.txt') or File.link('README', 'README.txt')
   hoe = Hoe.new(name, version) do |p|
     p.author = spec.author
     p.description = spec.description

data/ext/fast_xs/fast_xs.c

@@ -1,53 +1,13 @@
-#define VERSION	"0.1"
-
 #include <ruby.h>
 #include <assert.h>
 #include <sys/time.h>
 #include <sys/resource.h>
 #include "ruby_1_9_compat.h"
-
-/* I don't trust ctype.h when it comes to locale-independence: */
-static __inline__ int is_hex(const int x)
-{
-	return (((x) >= '0' && (x) <= '9') ||
-                ((x) >= 'a' && (x) <= 'f') ||
-		((x) >= 'A' && (x) <= 'F'));
-}
-
-static __inline__ int xtoupper(const int x)
-{
-	return (x >= 'a' && x <= 'f') ? (x & ~0x20) : x;
-}
-
-static __inline__ int hexchar_to_int(const int x)
-{
-	return (x < 'A') ? (x - '0') : (xtoupper(x) - 'A' + 10);
-}
-
-static __inline__ int hexpair_to_int(const int x1, const int x2)
-{
-	return ((hexchar_to_int(x1) << 4) | hexchar_to_int(x2));
-}
+#include "fast_xs_type.h"
+#include "gcc.h"
 
 static ID unpack_id;
 static VALUE U_fmt, C_fmt;
-static rlim_t alloca_limit = 4096; /* very small default */
-
-#define xs_alloc(size) \
-	unlikely((size) > alloca_limit) ? malloc(size) : alloca(size)
-
-#define xs_free(ptr, size) \
-	do { if (unlikely((size) > alloca_limit)) free(ptr); } while (0)
-
-/* give GCC hints for better branch prediction
- * (we layout branches so that ASCII characters are handled faster) */
-#if defined(__GNUC__) && (__GNUC__ >= 3)
-#  define likely(x)		__builtin_expect (!!(x), 1)
-#  define unlikely(x)		__builtin_expect (!!(x), 0)
-#else
-#  define unlikely(x)		(x)
-#  define likely(x)		(x)
-#endif
 
 /* pass-through certain characters for CP-1252 */
 #define p(x) (x-128)
@@ -172,7 +132,7 @@ static VALUE fast_xs(VALUE self)
 {
 	long i;
 	VALUE array;
-	char *s, *c;
+	char *c;
 	size_t s_len;
 	VALUE *tmp;
 	VALUE rv;
@@ -197,15 +157,12 @@ static VALUE fast_xs(VALUE self)
 			s_len += bytes_for(n) - 1;
 	}
 
-	c = s = unlikely(s_len > alloca_limit) ? malloc(s_len) : alloca(s_len);
+	rv = rb_str_new(NULL, s_len);
+	c = RSTRING_PTR(rv);
 
 	for (tmp = RARRAY_PTR(array), i = RARRAY_LEN(array); --i >= 0; tmp++)
 		c += escape(c, NUM2INT(*tmp));
 
-	rv = rb_str_new(s, s_len);
-
-	xs_free(s, s_len);
-
 	return rv;
 }
 
@@ -232,7 +189,8 @@ static VALUE fast_xs_html(VALUE self)
 			new_len += (sizeof("&quot;") - 2);
 	}
 
-	new_str = xs_alloc(new_len);
+	rv = rb_str_new(NULL, new_len);
+	new_str = RSTRING_PTR(rv);
 
 #define append_const(buf, x) do { \
 	buf = memcpy(buf, x, sizeof(x) - 1) + sizeof(x) - 1; \
@@ -253,10 +211,6 @@ static VALUE fast_xs_html(VALUE self)
 
 #undef append_const
 
-	rv = rb_str_new(new_str - new_len, new_len);
-
-	xs_free(new_str - new_len, new_len);
-
 	return rv;
 }
 
@@ -280,7 +234,8 @@ static inline VALUE _xs_uri_encode(VALUE self, const unsigned int space_to_plus)
 		new_len += 2;
 	}
 
-	new_str = xs_alloc(new_len);
+	rv = rb_str_new(NULL, new_len);
+	new_str = RSTRING_PTR(rv);
 
 	for (s = RSTRING_PTR(self), i = RSTRING_LEN(self); --i >= 0; ++s) {
 		if (likely(CGI_URI_OK(*s)))
@@ -296,10 +251,6 @@ static inline VALUE _xs_uri_encode(VALUE self, const unsigned int space_to_plus)
 		}
 	}
 
-	rv = rb_str_new(new_str - new_len, new_len);
-
-	xs_free(new_str - new_len, new_len);
-
 	return rv;
 }
 
@@ -341,7 +292,9 @@ static VALUE _uxs_uri(VALUE self, const unsigned int plus_to_space)
 		}
 	}
 
-	new_str = xs_alloc(new_len);
+	rv = rb_str_new(NULL, new_len);
+	new_str = RSTRING_PTR(rv);
+
 	for (s = RSTRING_PTR(self), i = RSTRING_LEN(self);
 	     --i >= 0;
 	     ++s, ++new_str) {
@@ -357,10 +310,6 @@ static VALUE _uxs_uri(VALUE self, const unsigned int plus_to_space)
 			*new_str = *s;
 	}
 
-	rv = rb_str_new(new_str - new_len, new_len);
-
-	xs_free(s, new_len);
-
 	return rv;
 }
 
@@ -371,14 +320,8 @@ static VALUE fast_uxs_cgi(VALUE self)
 
 void Init_fast_xs(void)
 {
-	struct rlimit rlim;
-
 	assert(cp_1252[159 - 128] == 376); /* just in case I skipped a line */
 
-	/* fairly conservative stack estimation IMHO... */
-	if (!getrlimit(RLIMIT_STACK, &rlim) && (rlim.rlim_cur > 0x80000))
-		alloca_limit = rlim.rlim_cur - (rlim.rlim_cur / 16);
-
 	unpack_id = rb_intern("unpack");
 	U_fmt = rb_str_new("U*", 2);
 	C_fmt = rb_str_new("C*", 2);

data/ext/fast_xs/fast_xs_type.h

@@ -0,0 +1,28 @@
+#ifndef FAST_XS_TYPE_H
+#define  FAST_XS_TYPE_H
+
+/* I don't trust ctype.h when it comes to locale-independence: */
+
+static __inline__ int is_hex(const int x)
+{
+	return (((x) >= '0' && (x) <= '9') ||
+	       ((x) >= 'a' && (x) <= 'f') ||
+	       ((x) >= 'A' && (x) <= 'F'));
+}
+
+static __inline__ int xtoupper(const int x)
+{
+	return (x >= 'a' && x <= 'f') ? (x & ~0x20) : x;
+}
+
+static __inline__ int hexchar_to_int(const int x)
+{
+	return (x < 'A') ? (x - '0') : (xtoupper(x) - 'A' + 10);
+}
+
+static __inline__ int hexpair_to_int(const int x1, const int x2)
+{
+	return ((hexchar_to_int(x1) << 4) | hexchar_to_int(x2));
+}
+
+#endif /* FAST_XS_TYPE_H */

data/ext/fast_xs/gcc.h

@@ -0,0 +1,14 @@
+#ifndef GCC_H
+#define GCC_H
+
+/* give GCC hints for better branch prediction
+ * (we layout branches so that ASCII characters are handled faster) */
+#if defined(__GNUC__) && (__GNUC__ >= 3)
+#  define likely(x)		__builtin_expect (!!(x), 1)
+#  define unlikely(x)		__builtin_expect (!!(x), 0)
+#else
+#  define unlikely(x)		(x)
+#  define likely(x)		(x)
+#endif
+
+#endif /* GCC_H */

data/test/test_cgi_class_overrides.rb

@@ -36,6 +36,7 @@ class TestCgiClassOverrides < Test::Unit::TestCase
     assert_equal ' ', CGI::unescape('+')
     assert_equal '+', CGI::unescape('%2B')
     assert_equal ',', CGI::unescape('%2C')
+    assert_equal '%', CGI::unescape('%')
     assert_equal 'hello-world', CGI::unescape('hello-world')
     assert_equal 'H3LL0 W0RLD', CGI::unescape('H3LL0+W0RLD')
   end

metadata

@@ -3,8 +3,8 @@ rubygems_version: 0.9.4.7
 specification_version: 2
 name: fast_xs
 version: !ruby/object:Gem::Version 
-  version: "0.5"
-date: 2008-01-07 00:00:00 -08:00
+  version: "0.6"
+date: 2008-02-20 00:00:00 -08:00
 summary: excessively fast escaping
 require_paths: 
 - lib/i486-linux
@@ -48,6 +48,8 @@ files:
 - ext/fast_xs/fast_xs.c
 - ext/fast_xs/extconf.rb
 - ext/fast_xs/ruby_1_9_compat.h
+- ext/fast_xs/fast_xs_type.h
+- ext/fast_xs/gcc.h
 test_files: []
 
 rdoc_options: