Performace patch for ruby-1.9.3-p327 detailed

00-description.md

This are parts of falcon patch for ruby-1.9.3-p327

• 01-backport-speedup-require.diff - backport of Greg Price's patch for speedup require.

  For a long time my patch were famous cause of require speedup. ruby-core prefers slightly simpler patches from Greg Price, and I wish not to compete with ruby-core, so that I just backport accepted changes.

• 02-st_opt.diff - speedup of Hash

  This patch contains:

  1. backport of accepted patch which compacts small hashes
  2. ultrapacking for hashes of size 1
  3. slab allocation for hash items

• 03-sparse_array.diff - speedup of method/constant/instance variable lookup

  This patch introduce new hash structure instead of st_table for storing Constants, Methods and instance variable offsets. This structure is more compact and suffers less from collisions that st_table.

• 04-ary_queue.diff - make Array really suitable as a queue

  While there were a step in this direction in a ruby, this patch pushed it till the end. Now push/shift pattern do not suffers from big queue size and high interleaving of push/shift.

• 05-backport-gc.diff - backport COW-friendly GC

  Well, I update this patch so people could decide, worth it something or not. I were disappointed with it cause it doesn't save much memory with big application and tuned garbage collector.

• 06-st_opt_sparse_array.diff - premerged 02 and 03

falcon.diff is a merge of 01, 04 and 06

falcon-gc.diff is a merge of falcon.diff and 05

01-backport-speedup-require.diff

diff --git a/array.c b/array.c
index e427cb3..cc15660 100644
--- a/array.c
+++ b/array.c
@@ -295,6 +295,33 @@ rb_ary_frozen_p(VALUE ary)
     return Qfalse;
 }
 
+/* This can be used to take a snapshot of an array (with
+   e.g. rb_ary_replace) and check later whether the array has been
+   modified from the snapshot.  The snapshot is cheap, though if
+   something does modify the array it will pay the cost of copying
+   it. */
+VALUE
+rb_ary_dup_of_p(VALUE ary1, VALUE ary2)
+{
+    VALUE *p1, *p2;
+    long len = RARRAY_LEN(ary1);
+
+    if (len != RARRAY_LEN(ary2)) return Qfalse;
+
+    p1 = RARRAY_PTR(ary1);
+    p2 = RARRAY_PTR(ary2);
+
+    if (ARY_EMBED_P(ary1) && ARY_EMBED_P(ary2)) {
+        for (; len; len--, p1++, p2++) {
+            if (*p1 != *p2) return Qfalse;
+        }
+        return Qtrue;
+    }
+
+    if (p1 == p2) return Qtrue;
+    return Qfalse;
+}
+
 static VALUE
 ary_alloc(VALUE klass)
 {
diff --git a/file.c b/file.c
index c1db6d7..3f465e5 100644
--- a/file.c
+++ b/file.c
@@ -148,40 +148,60 @@ file_path_convert(VALUE name)
     return name;
 }
 
-static VALUE
-rb_get_path_check(VALUE obj, int level)
+static rb_encoding *
+check_path_encoding(VALUE str)
+{
+    rb_encoding *enc = rb_enc_get(str);
+    if (!rb_enc_asciicompat(enc)) {
+	rb_raise(rb_eEncCompatError, "path name must be ASCII-compatible (%s): %s",
+		 rb_enc_name(enc), RSTRING_PTR(rb_str_inspect(str)));
+    }
+    return enc;
+}
+
+VALUE
+rb_get_path_check_to_string(VALUE obj, int level)
 {
     VALUE tmp;
     ID to_path;
-    rb_encoding *enc;
 
     if (insecure_obj_p(obj, level)) {
 	rb_insecure_operation();
     }
 
+    if (RB_TYPE_P(obj, T_STRING)) {
+	return obj;
+    }
     CONST_ID(to_path, "to_path");
     tmp = rb_check_funcall(obj, to_path, 0, 0);
     if (tmp == Qundef) {
 	tmp = obj;
     }
     StringValue(tmp);
+    return tmp;
+}
 
+VALUE
+rb_get_path_check_convert(VALUE obj, VALUE tmp, int level)
+{
     tmp = file_path_convert(tmp);
     if (obj != tmp && insecure_obj_p(tmp, level)) {
 	rb_insecure_operation();
     }
-    enc = rb_enc_get(tmp);
-    if (!rb_enc_asciicompat(enc)) {
-	tmp = rb_str_inspect(tmp);
-	rb_raise(rb_eEncCompatError, "path name must be ASCII-compatible (%s): %s",
-		 rb_enc_name(enc), RSTRING_PTR(tmp));
-    }
 
+    check_path_encoding(tmp);
     StringValueCStr(tmp);
 
     return rb_str_new4(tmp);
 }
 
+static VALUE
+rb_get_path_check(VALUE obj, int level)
+{
+    VALUE tmp = rb_get_path_check_to_string(obj, level);
+    return rb_get_path_check_convert(obj, tmp, level);
+}
+
 VALUE
 rb_get_path_no_checksafe(VALUE obj)
 {
@@ -3249,7 +3269,6 @@ rb_file_expand_path(VALUE fname, VALUE dname)
 VALUE
 rb_file_expand_path_fast(VALUE fname, VALUE dname)
 {
-    check_expand_path_args(fname, dname);
     return rb_file_expand_path_internal(fname, dname, 0, 0, EXPAND_PATH_BUFFER());
 }
 
@@ -5237,7 +5256,7 @@ rb_find_file_ext_safe(VALUE *filep, const char *const *ext, int safe_level)
 	rb_raise(rb_eSecurityError, "loading from non-absolute path %s", f);
     }
 
-    RB_GC_GUARD(load_path) = rb_get_load_path();
+    RB_GC_GUARD(load_path) = rb_get_expanded_load_path();
     if (!load_path) return 0;
 
     fname = rb_str_dup(*filep);
@@ -5302,7 +5321,7 @@ rb_find_file_safe(VALUE path, int safe_level)
 	rb_raise(rb_eSecurityError, "loading from non-absolute path %s", f);
     }
 
-    RB_GC_GUARD(load_path) = rb_get_load_path();
+    RB_GC_GUARD(load_path) = rb_get_expanded_load_path();
     if (load_path) {
 	long i;
 
diff --git a/hash.c b/hash.c
index fbd8237..8423558 100644
--- a/hash.c
+++ b/hash.c
@@ -1087,7 +1087,7 @@ clear_i(VALUE key, VALUE value, VALUE dummy)
  *
  */
 
-static VALUE
+VALUE
 rb_hash_clear(VALUE hash)
 {
     rb_hash_modify_check(hash);
diff --git a/include/ruby/intern.h b/include/ruby/intern.h
index 927b536..0781de8 100644
--- a/include/ruby/intern.h
+++ b/include/ruby/intern.h
@@ -56,6 +56,7 @@ VALUE rb_ary_tmp_new(long);
 void rb_ary_free(VALUE);
 void rb_ary_modify(VALUE);
 VALUE rb_ary_freeze(VALUE);
+VALUE rb_ary_dup_of_p(VALUE, VALUE);
 VALUE rb_ary_aref(int, VALUE*, VALUE);
 VALUE rb_ary_subseq(VALUE, long, long);
 void rb_ary_store(VALUE, long, VALUE);
@@ -440,6 +441,7 @@ VALUE rb_hash_lookup(VALUE, VALUE);
 VALUE rb_hash_lookup2(VALUE, VALUE, VALUE);
 VALUE rb_hash_fetch(VALUE, VALUE);
 VALUE rb_hash_aset(VALUE, VALUE, VALUE);
+VALUE rb_hash_clear(VALUE);
 VALUE rb_hash_delete_if(VALUE);
 VALUE rb_hash_delete(VALUE,VALUE);
 typedef VALUE rb_hash_update_func(VALUE newkey, VALUE oldkey, VALUE value);
diff --git a/internal.h b/internal.h
index 59c9284..f5af903 100644
--- a/internal.h
+++ b/internal.h
@@ -94,6 +94,8 @@ VALUE rb_home_dir(const char *user, VALUE result);
 VALUE rb_realpath_internal(VALUE basedir, VALUE path, int strict);
 VALUE rb_file_expand_path_fast(VALUE, VALUE);
 VALUE rb_file_expand_path_internal(VALUE, VALUE, int, int, VALUE);
+VALUE rb_get_path_check_to_string(VALUE, int);
+VALUE rb_get_path_check_convert(VALUE, VALUE, int);
 void Init_File(void);
 
 #ifdef _WIN32
@@ -119,6 +121,7 @@ VALUE rb_iseq_clone(VALUE iseqval, VALUE newcbase);
 
 /* load.c */
 VALUE rb_get_load_path(void);
+VALUE rb_get_expanded_load_path(void);
 
 /* math.c */
 VALUE rb_math_atan2(VALUE, VALUE);
diff --git a/load.c b/load.c
index 163ec4c..68caebc 100644
--- a/load.c
+++ b/load.c
@@ -34,21 +34,120 @@ rb_get_load_path(void)
     return load_path;
 }
 
-VALUE
-rb_get_expanded_load_path(void)
+enum expand_type {
+    EXPAND_ALL,
+    EXPAND_RELATIVE,
+    EXPAND_HOME,
+    EXPAND_NON_CACHE
+};
+
+/* Construct expanded load path and store it to cache.
+   We rebuild load path partially if the cache is invalid.
+   We don't cache non string object and expand it every time. We ensure that
+   string objects in $LOAD_PATH are frozen.
+ */
+static void
+rb_construct_expanded_load_path(int type, int *has_relative, int *has_non_cache)
 {
-    VALUE load_path = rb_get_load_path();
+    rb_vm_t *vm = GET_VM();
+    VALUE load_path = vm->load_path;
+    VALUE expanded_load_path = vm->expanded_load_path;
     VALUE ary;
     long i;
+    int level = rb_safe_level();
 
     ary = rb_ary_new2(RARRAY_LEN(load_path));
     for (i = 0; i < RARRAY_LEN(load_path); ++i) {
-	VALUE path = rb_file_expand_path_fast(RARRAY_PTR(load_path)[i], Qnil);
-	rb_str_freeze(path);
-	rb_ary_push(ary, path);
+	VALUE path, as_str, expanded_path;
+	int is_string, non_cache;
+	char *as_cstr;
+	as_str = path = RARRAY_PTR(load_path)[i];
+	is_string = RB_TYPE_P(path, T_STRING) ? 1 : 0;
+	non_cache = !is_string ? 1 : 0;
+	as_str = rb_get_path_check_to_string(path, level);
+	as_cstr = RSTRING_PTR(as_str);
+
+	if (!non_cache) {
+	    if ((type == EXPAND_RELATIVE &&
+		    rb_is_absolute_path(as_cstr)) ||
+		(type == EXPAND_HOME &&
+		    (!as_cstr[0] || as_cstr[0] != '~')) ||
+		(type == EXPAND_NON_CACHE)) {
+		    /* Use cached expanded path. */
+		    rb_ary_push(ary, RARRAY_PTR(expanded_load_path)[i]);
+		    continue;
+	    }
+	}
+	if (!*has_relative && !rb_is_absolute_path(as_cstr))
+	    *has_relative = 1;
+	if (!*has_non_cache && non_cache)
+	    *has_non_cache = 1;
+	/* Freeze only string object. We expand other objects every time. */
+	if (is_string)
+	    rb_str_freeze(path);
+	as_str = rb_get_path_check_convert(path, as_str, level);
+	expanded_path = rb_file_expand_path_fast(as_str, Qnil);
+	rb_str_freeze(expanded_path);
+	rb_ary_push(ary, expanded_path);
     }
     rb_obj_freeze(ary);
-    return ary;
+    vm->expanded_load_path = ary;
+    rb_ary_replace(vm->load_path_snapshot, vm->load_path);
+}
+
+static VALUE
+load_path_getcwd(void)
+{
+    char *cwd = my_getcwd();
+    VALUE cwd_str = rb_filesystem_str_new_cstr(cwd);
+    xfree(cwd);
+    return cwd_str;
+}
+
+VALUE
+rb_get_expanded_load_path(void)
+{
+    rb_vm_t *vm = GET_VM();
+    const VALUE non_cache = Qtrue;
+
+    if (!rb_ary_dup_of_p(vm->load_path_snapshot, vm->load_path)) {
+	/* The load path was modified. Rebuild the expanded load path. */
+	int has_relative = 0, has_non_cache = 0;
+	rb_construct_expanded_load_path(EXPAND_ALL, &has_relative, &has_non_cache);
+	if (has_relative) {
+	    vm->load_path_check_cache = load_path_getcwd();
+	}
+	else if (has_non_cache) {
+	    /* Non string object. */
+	    vm->load_path_check_cache = non_cache;
+	}
+	else {
+	    vm->load_path_check_cache = 0;
+	}
+    }
+    else if (vm->load_path_check_cache == non_cache) {
+	int has_relative = 1, has_non_cache = 1;
+	/* Expand only non-cacheable objects. */
+	rb_construct_expanded_load_path(EXPAND_NON_CACHE,
+					&has_relative, &has_non_cache);
+    }
+    else if (vm->load_path_check_cache) {
+	int has_relative = 1, has_non_cache = 1;
+	VALUE cwd = load_path_getcwd();
+	if (!rb_str_equal(vm->load_path_check_cache, cwd)) {
+	    /* Current working directory or filesystem encoding was changed.
+	       Expand relative load path and non-cacheable objects again. */
+	    vm->load_path_check_cache = cwd;
+	    rb_construct_expanded_load_path(EXPAND_RELATIVE,
+					    &has_relative, &has_non_cache);
+	}
+	else {
+	    /* Expand only tilde (User HOME) and non-cacheable objects. */
+	    rb_construct_expanded_load_path(EXPAND_HOME,
+					    &has_relative, &has_non_cache);
+	}
+    }
+    return vm->expanded_load_path;
 }
 
 static VALUE
@@ -63,12 +162,121 @@ get_loaded_features(void)
     return GET_VM()->loaded_features;
 }
 
+static void
+reset_loaded_features_snapshot(void)
+{
+    rb_vm_t *vm = GET_VM();
+    rb_ary_replace(vm->loaded_features_snapshot, vm->loaded_features);
+}
+
+static VALUE
+get_loaded_features_index_raw(void)
+{
+    return GET_VM()->loaded_features_index;
+}
+
 static st_table *
 get_loading_table(void)
 {
     return GET_VM()->loading_table;
 }
 
+static void
+features_index_add_single(VALUE short_feature, VALUE offset)
+{
+    VALUE features_index, this_feature_index;
+    features_index = get_loaded_features_index_raw();
+    if ((this_feature_index = rb_hash_lookup(features_index, short_feature)) == Qnil) {
+	this_feature_index = rb_ary_new();
+	rb_hash_aset(features_index, short_feature, this_feature_index);
+    }
+    rb_ary_push(this_feature_index, offset);
+}
+
+/* Add to the loaded-features index all the required entries for
+   `feature`, located at `offset` in $LOADED_FEATURES.  We add an
+   index entry at each string `short_feature` for which
+     feature == "#{prefix}#{short_feature}#{e}"
+   where `e` is empty or matches %r{^\.[^./]*$}, and `prefix` is empty
+   or ends in '/'.  This maintains the invariant that `rb_feature_p()`
+   relies on for its fast lookup.
+*/
+static void
+features_index_add(VALUE feature, VALUE offset)
+{
+    VALUE short_feature;
+    const char *feature_str, *feature_end, *ext, *p;
+
+    feature_str = StringValuePtr(feature);
+    feature_end = feature_str + RSTRING_LEN(feature);
+
+    for (ext = feature_end; ext > feature_str; ext--)
+      if (*ext == '.' || *ext == '/')
+	break;
+    if (*ext != '.')
+      ext = NULL;
+    /* Now `ext` points to the only string matching %r{^\.[^./]*$} that is
+       at the end of `feature`, or is NULL if there is no such string. */
+
+    p = ext ? ext : feature_end;
+    while (1) {
+	p--;
+	while (p >= feature_str && *p != '/')
+	    p--;
+	if (p < feature_str)
+	    break;
+	/* Now *p == '/'.  We reach this point for every '/' in `feature`. */
+	short_feature = rb_str_substr(feature, p + 1 - feature_str, feature_end - p - 1);
+	features_index_add_single(short_feature, offset);
+	if (ext) {
+	    short_feature = rb_str_substr(feature, p + 1 - feature_str, ext - p - 1);
+	    features_index_add_single(short_feature, offset);
+	}
+    }
+    features_index_add_single(feature, offset);
+    if (ext) {
+	short_feature = rb_str_substr(feature, 0, ext - feature_str);
+	features_index_add_single(short_feature, offset);
+    }
+}
+
+static VALUE
+get_loaded_features_index(void)
+{
+    VALUE features;
+    int i;
+    rb_vm_t *vm = GET_VM();
+
+    if (!rb_ary_dup_of_p(vm->loaded_features_snapshot, vm->loaded_features)) {
+	/* The sharing was broken; something (other than us in rb_provide_feature())
+	   modified loaded_features.  Rebuild the index. */
+	rb_hash_clear(vm->loaded_features_index);
+	features = vm->loaded_features;
+	for (i = 0; i < RARRAY_LEN(features); i++) {
+	    VALUE entry, as_str;
+	    as_str = entry = rb_ary_entry(features, i);
+	    StringValue(as_str);
+	    if (as_str != entry)
+		rb_ary_store(features, i, as_str);
+	    rb_str_freeze(as_str);
+	    features_index_add(as_str, INT2FIX(i));
+	}
+	reset_loaded_features_snapshot();
+    }
+    return vm->loaded_features_index;
+}
+
+/* This searches `load_path` for a value such that
+     name == "#{load_path[i]}/#{feature}"
+   if `feature` is a suffix of `name`, or otherwise
+     name == "#{load_path[i]}/#{feature}#{ext}"
+   for an acceptable string `ext`.  It returns
+   `load_path[i].to_str` if found, else 0.
+
+   If type is 's', then `ext` is acceptable only if IS_DLEXT(ext);
+   if 'r', then only if IS_RBEXT(ext); otherwise `ext` may be absent
+   or have any value matching `%r{^\.[^./]*$}`.
+*/
 static VALUE
 loaded_feature_path(const char *name, long vlen, const char *feature, long len,
 		    int type, VALUE load_path)
@@ -88,23 +296,22 @@ loaded_feature_path(const char *name, long vlen, const char *feature, long len,
 	    return 0;
 	plen = e - name - len - 1;
     }
+    if (type == 's' && !IS_DLEXT(&name[plen+len+1])
+     || type == 'r' && !IS_RBEXT(&name[plen+len+1])
+     || name[plen] != '/') {
+       return 0;
+    }
+    /* Now name == "#{prefix}/#{feature}#{ext}" where ext is acceptable
+       (possibly empty) and prefix is some string of length plen. */
+
     for (i = 0; i < RARRAY_LEN(load_path); ++i) {
 	VALUE p = RARRAY_PTR(load_path)[i];
 	const char *s = StringValuePtr(p);
 	long n = RSTRING_LEN(p);
 
-	if (n != plen ) continue;
-	if (n && (strncmp(name, s, n) || name[n] != '/')) continue;
-	switch (type) {
-	  case 's':
-	    if (IS_DLEXT(&name[n+len+1])) return p;
-	    break;
-	  case 'r':
-	    if (IS_RBEXT(&name[n+len+1])) return p;
-	    break;
-	  default:
-	    return p;
-	}
+	if (n != plen) continue;
+	if (n && strncmp(name, s, n)) continue;
+	return p;
     }
     return 0;
 }
@@ -132,7 +339,7 @@ loaded_feature_path_i(st_data_t v, st_data_t b, st_data_t f)
 static int
 rb_feature_p(const char *feature, const char *ext, int rb, int expanded, const char **fn)
 {
-    VALUE v, features, p, load_path = 0;
+    VALUE features, features_index, feature_val, this_feature_index, v, p, load_path = 0;
     const char *f, *e;
     long i, len, elen, n;
     st_table *loading_tbl;
@@ -151,8 +358,39 @@ rb_feature_p(const char *feature, const char *ext, int rb, int expanded, const c
 	type = 0;
     }
     features = get_loaded_features();
-    for (i = 0; i < RARRAY_LEN(features); ++i) {
-	v = RARRAY_PTR(features)[i];
+    features_index = get_loaded_features_index();
+
+    feature_val = rb_str_new(feature, len);
+    this_feature_index = rb_hash_lookup(features_index, feature_val);
+    /* We search `features` for an entry such that either
+         "#{features[i]}" == "#{load_path[j]}/#{feature}#{e}"
+       for some j, or
+         "#{features[i]}" == "#{feature}#{e}"
+       Here `e` is an "allowed" extension -- either empty or one
+       of the extensions accepted by IS_RBEXT, IS_SOEXT, or
+       IS_DLEXT.  Further, if `ext && rb` then `IS_RBEXT(e)`,
+       and if `ext && !rb` then `IS_SOEXT(e) || IS_DLEXT(e)`.
+
+       If `expanded`, then only the latter form (without load_path[j])
+       is accepted.  Otherwise either form is accepted, *unless* `ext`
+       is false and an otherwise-matching entry of the first form is
+       preceded by an entry of the form
+         "#{features[i2]}" == "#{load_path[j2]}/#{feature}#{e2}"
+       where `e2` matches %r{^\.[^./]*$} but is not an allowed extension.
+       After a "distractor" entry of this form, only entries of the
+       form "#{feature}#{e}" are accepted.
+
+       In `rb_provide_feature()` and `get_loaded_features_index()` we
+       maintain an invariant that the array `this_feature_index` will
+       point to every entry in `features` which has the form
+         "#{prefix}#{feature}#{e}"
+       where `e` is empty or matches %r{^\.[^./]*$}, and `prefix` is empty
+       or ends in '/'.  This includes both match forms above, as well
+       as any distractors, so we may ignore all other entries in `features`.
+     */
+    for (i = 0; this_feature_index != Qnil && i < RARRAY_LEN(this_feature_index); i++) {
+	long index = FIX2LONG(rb_ary_entry(this_feature_index, i));
+	v = RARRAY_PTR(features)[index];
 	f = StringValuePtr(v);
 	if ((n = RSTRING_LEN(v)) < len) continue;
 	if (strncmp(f, feature, len) != 0) {
@@ -175,6 +413,7 @@ rb_feature_p(const char *feature, const char *ext, int rb, int expanded, const c
 	    return 'r';
 	}
     }
+
     loading_tbl = get_loading_table();
     if (loading_tbl) {
 	f = 0;
@@ -183,7 +422,7 @@ rb_feature_p(const char *feature, const char *ext, int rb, int expanded, const c
 	    fs.name = feature;
 	    fs.len = len;
 	    fs.type = type;
-	    fs.load_path = load_path ? load_path : rb_get_load_path();
+	    fs.load_path = load_path ? load_path : rb_get_expanded_load_path();
 	    fs.result = 0;
 	    st_foreach(loading_tbl, loaded_feature_path_i, (st_data_t)&fs);
 	    if ((f = fs.result) != 0) {
@@ -233,7 +472,7 @@ rb_feature_provided(const char *feature, const char **loading)
 
     if (*feature == '.' &&
 	(feature[1] == '/' || strncmp(feature+1, "./", 2) == 0)) {
-	fullpath = rb_file_expand_path_fast(rb_str_new2(feature), Qnil);
+	fullpath = rb_file_expand_path_fast(rb_get_path(rb_str_new2(feature)), Qnil);
 	feature = RSTRING_PTR(fullpath);
     }
     if (ext && !strchr(ext, '/')) {
@@ -254,11 +493,18 @@ rb_feature_provided(const char *feature, const char **loading)
 static void
 rb_provide_feature(VALUE feature)
 {
-    if (OBJ_FROZEN(get_loaded_features())) {
+    VALUE features;
+
+    features = get_loaded_features();
+    if (OBJ_FROZEN(features)) {
 	rb_raise(rb_eRuntimeError,
 		 "$LOADED_FEATURES is frozen; cannot append feature");
     }
-    rb_ary_push(get_loaded_features(), feature);
+    rb_str_freeze(feature);
+
+    rb_ary_push(features, feature);
+    features_index_add(feature, INT2FIX(RARRAY_LEN(features)-1));
+    reset_loaded_features_snapshot();
 }
 
 void
@@ -774,10 +1020,15 @@ Init_load()
     rb_alias_variable(rb_intern("$-I"), id_load_path);
     rb_alias_variable(rb_intern("$LOAD_PATH"), id_load_path);
     vm->load_path = rb_ary_new();
+    vm->expanded_load_path = rb_ary_new();
+    vm->load_path_snapshot = rb_ary_new();
+    vm->load_path_check_cache = 0;
 
     rb_define_virtual_variable("$\"", get_loaded_features, 0);
     rb_define_virtual_variable("$LOADED_FEATURES", get_loaded_features, 0);
     vm->loaded_features = rb_ary_new();
+    vm->loaded_features_snapshot = rb_ary_new();
+    vm->loaded_features_index = rb_hash_new();
 
     rb_define_global_function("load", rb_f_load, -1);
     rb_define_global_function("require", rb_f_require, 1);
diff --git a/ruby.c b/ruby.c
index 3ddd96c..7ffc78e 100644
--- a/ruby.c
+++ b/ruby.c
@@ -1366,7 +1366,8 @@ process_options(int argc, char **argv, struct cmdline_options *opt)
 	long i;
 	VALUE load_path = GET_VM()->load_path;
 	for (i = 0; i < RARRAY_LEN(load_path); ++i) {
-	    rb_enc_associate(RARRAY_PTR(load_path)[i], lenc);
+	    RARRAY_PTR(load_path)[i] =
+		rb_enc_associate(rb_str_dup(RARRAY_PTR(load_path)[i]), lenc);
 	}
     }
     if (!(opt->disable & DISABLE_BIT(gems))) {
diff --git a/test/ruby/test_require.rb b/test/ruby/test_require.rb
index 58a9ee2..ec75096 100644
--- a/test/ruby/test_require.rb
+++ b/test/ruby/test_require.rb
@@ -356,4 +356,114 @@ class TestRequire < Test::Unit::TestCase
     $:.replace(loadpath)
     $".replace(features)
   end
+
+  def test_require_changed_current_dir
+    bug7158 = '[ruby-core:47970]'
+    Dir.mktmpdir {|tmp|
+      Dir.chdir(tmp) {
+        Dir.mkdir("a")
+        Dir.mkdir("b")
+        open(File.join("a", "foo.rb"), "w") {}
+        open(File.join("b", "bar.rb"), "w") {|f|
+          f.puts "p :ok"
+        }
+        assert_in_out_err([], <<-INPUT, %w(:ok), [], bug7158)
+          $: << "."
+          Dir.chdir("a")
+          require "foo"
+          Dir.chdir("../b")
+          p :ng unless require "bar"
+          Dir.chdir("..")
+          p :ng if require "b/bar"
+        INPUT
+      }
+    }
+  end
+
+  def test_require_not_modified_load_path
+    bug7158 = '[ruby-core:47970]'
+    Dir.mktmpdir {|tmp|
+      Dir.chdir(tmp) {
+        open("foo.rb", "w") {}
+        assert_in_out_err([], <<-INPUT, %w(:ok), [], bug7158)
+          a = Object.new
+          def a.to_str
+            "#{tmp}"
+          end
+          $: << a
+          require "foo"
+          last_path = $:.pop
+          p :ok if last_path == a && last_path.class == Object
+        INPUT
+      }
+    }
+  end
+
+  def test_require_changed_home
+    bug7158 = '[ruby-core:47970]'
+    Dir.mktmpdir {|tmp|
+      Dir.chdir(tmp) {
+        open("foo.rb", "w") {}
+        Dir.mkdir("a")
+        open(File.join("a", "bar.rb"), "w") {}
+        assert_in_out_err([], <<-INPUT, %w(:ok), [], bug7158)
+          $: << '~'
+          ENV['HOME'] = "#{tmp}"
+          require "foo"
+          ENV['HOME'] = "#{tmp}/a"
+          p :ok if require "bar"
+        INPUT
+      }
+    }
+  end
+
+  def test_require_to_path_redefined_in_load_path
+    bug7158 = '[ruby-core:47970]'
+    Dir.mktmpdir {|tmp|
+      Dir.chdir(tmp) {
+        open("foo.rb", "w") {}
+        assert_in_out_err(["RUBYOPT"=>nil], <<-INPUT, %w(:ok), [], bug7158)
+          a = Object.new
+          def a.to_path
+            "bar"
+          end
+          $: << a
+          begin
+            require "foo"
+            p :ng
+          rescue LoadError
+          end
+          def a.to_path
+            "#{tmp}"
+          end
+          p :ok if require "foo"
+        INPUT
+      }
+    }
+  end
+
+  def test_require_to_str_redefined_in_load_path
+    bug7158 = '[ruby-core:47970]'
+    Dir.mktmpdir {|tmp|
+      Dir.chdir(tmp) {
+        open("foo.rb", "w") {}
+        assert_in_out_err(["RUBYOPT"=>nil], <<-INPUT, %w(:ok), [], bug7158)
+          a = Object.new
+          def a.to_str
+            "foo"
+          end
+          $: << a
+          begin
+            require "foo"
+            p :ng
+          rescue LoadError
+          end
+          def a.to_str
+            "#{tmp}"
+          end
+          p :ok if require "foo"
+        INPUT
+      }
+    }
+  end
 end
diff --git a/vm.c b/vm.c
index 63141ba..c187bb6 100644
--- a/vm.c
+++ b/vm.c
@@ -1578,7 +1578,12 @@ rb_vm_mark(void *ptr)
 	RUBY_MARK_UNLESS_NULL(vm->thgroup_default);
 	RUBY_MARK_UNLESS_NULL(vm->mark_object_ary);
 	RUBY_MARK_UNLESS_NULL(vm->load_path);
+	RUBY_MARK_UNLESS_NULL(vm->load_path_snapshot);
+	RUBY_MARK_UNLESS_NULL(vm->load_path_check_cache);
+	RUBY_MARK_UNLESS_NULL(vm->expanded_load_path);
 	RUBY_MARK_UNLESS_NULL(vm->loaded_features);
+	RUBY_MARK_UNLESS_NULL(vm->loaded_features_snapshot);
+	RUBY_MARK_UNLESS_NULL(vm->loaded_features_index);
 	RUBY_MARK_UNLESS_NULL(vm->top_self);
 	RUBY_MARK_UNLESS_NULL(vm->coverages);
 	rb_gc_mark_locations(vm->special_exceptions, vm->special_exceptions + ruby_special_error_count);
diff --git a/vm_core.h b/vm_core.h
index 60146f0..7b25806 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -298,7 +298,12 @@ typedef struct rb_vm_struct {
     /* load */
     VALUE top_self;
     VALUE load_path;
+    VALUE load_path_snapshot;
+    VALUE load_path_check_cache;
+    VALUE expanded_load_path;
     VALUE loaded_features;
+    VALUE loaded_features_snapshot;
+    VALUE loaded_features_index;
     struct st_table *loading_table;
 
     /* signal */

02-st_opt.diff

diff --git a/common.mk b/common.mk
index c9ef641..56e52b6 100644
--- a/common.mk
+++ b/common.mk
@@ -638,7 +638,8 @@ file.$(OBJEXT): {$(VPATH)}file.c $(RUBY_H_INCLUDES) {$(VPATH)}io.h \
 gc.$(OBJEXT): {$(VPATH)}gc.c $(RUBY_H_INCLUDES) {$(VPATH)}re.h \
   {$(VPATH)}regex.h $(ENCODING_H_INCLUDES) $(VM_CORE_H_INCLUDES) \
   {$(VPATH)}gc.h {$(VPATH)}io.h {$(VPATH)}eval_intern.h {$(VPATH)}util.h \
-  {$(VPATH)}debug.h {$(VPATH)}internal.h {$(VPATH)}constant.h
+  {$(VPATH)}debug.h {$(VPATH)}internal.h {$(VPATH)}constant.h \
+  {$(VPATH)}pool_alloc.inc.h {$(VPATH)}pool_alloc.h
 hash.$(OBJEXT): {$(VPATH)}hash.c $(RUBY_H_INCLUDES) {$(VPATH)}util.h \
   $(ENCODING_H_INCLUDES)
 inits.$(OBJEXT): {$(VPATH)}inits.c $(RUBY_H_INCLUDES) \
@@ -702,7 +703,7 @@ signal.$(OBJEXT): {$(VPATH)}signal.c $(RUBY_H_INCLUDES) \
   $(VM_CORE_H_INCLUDES) {$(VPATH)}debug.h
 sprintf.$(OBJEXT): {$(VPATH)}sprintf.c $(RUBY_H_INCLUDES) {$(VPATH)}re.h \
   {$(VPATH)}regex.h {$(VPATH)}vsnprintf.c $(ENCODING_H_INCLUDES)
-st.$(OBJEXT): {$(VPATH)}st.c $(RUBY_H_INCLUDES)
+st.$(OBJEXT): {$(VPATH)}st.c $(RUBY_H_INCLUDES) {$(VPATH)}pool_alloc.h
 strftime.$(OBJEXT): {$(VPATH)}strftime.c $(RUBY_H_INCLUDES) \
   {$(VPATH)}timev.h
 string.$(OBJEXT): {$(VPATH)}string.c $(RUBY_H_INCLUDES) {$(VPATH)}re.h \
diff --git a/configure.in b/configure.in
index b006a01..a3ccad7 100644
--- a/configure.in
+++ b/configure.in
@@ -1324,6 +1324,29 @@ main() {
 CFLAGS="$save_CFLAGS"])
 AC_DEFINE_UNQUOTED(GC_MARK_STACKFRAME_WORD, $rb_cv_gc_mark_stackframe_word)
 
+AS_CASE(["$target_os"],
+[openbsd*], [
+  AC_CACHE_CHECK(for heap align log on openbsd, rb_cv_page_size_log,
+    [rb_cv_page_size_log=no
+     for page_log in 12 13; do
+       AC_TRY_RUN([
+#include <math.h>
+#include <unistd.h>
+
+int
+main() {
+  if ((int)log2((double)sysconf(_SC_PAGESIZE)) != $page_log) return 1;
+  return 0;
+}
+       ],
+       rb_cv_page_size_log="$page_log"; break)
+     done])
+  if test $rb_cv_page_size_log != no; then
+    AC_DEFINE_UNQUOTED(HEAP_ALIGN_LOG, $rb_cv_page_size_log)
+  else
+    AC_DEFINE_UNQUOTED(HEAP_ALIGN_LOG, 12)
+  fi
+])
 
 dnl Checks for library functions.
 AC_TYPE_GETGROUPS
@@ -1424,7 +1447,8 @@ AC_CHECK_FUNCS(fmod killpg wait4 waitpid fork spawnv syscall __syscall chroot ge
 	      setsid telldir seekdir fchmod cosh sinh tanh log2 round\
 	      setuid setgid daemon select_large_fdset setenv unsetenv\
               mktime timegm gmtime_r clock_gettime gettimeofday poll ppoll\
-              pread sendfile shutdown sigaltstack dl_iterate_phdr)
+              pread sendfile shutdown sigaltstack dl_iterate_phdr\
+              dup3 pipe2 posix_memalign memalign)
 
 AC_CACHE_CHECK(for unsetenv returns a value, rb_cv_unsetenv_return_value,
   [AC_TRY_COMPILE([
diff --git a/ext/-test-/st/numhash/numhash.c b/ext/-test-/st/numhash/numhash.c
index e186cd4..53d9e1b 100644
--- a/ext/-test-/st/numhash/numhash.c
+++ b/ext/-test-/st/numhash/numhash.c
@@ -54,7 +54,7 @@ numhash_i(st_data_t key, st_data_t value, st_data_t arg, int error)
 static VALUE
 numhash_each(VALUE self)
 {
-    return st_foreach((st_table *)DATA_PTR(self), numhash_i, self) ? Qtrue : Qfalse;
+    return st_foreach_check((st_table *)DATA_PTR(self), numhash_i, self, 0) ? Qtrue : Qfalse;
 }
 
 void
diff --git a/gc.c b/gc.c
index e65d0ec..10ae77c 100644
--- a/gc.c
+++ b/gc.c
@@ -20,10 +20,12 @@
 #include "vm_core.h"
 #include "internal.h"
 #include "gc.h"
+#include "pool_alloc.h"
 #include "constant.h"
 #include <stdio.h>
 #include <setjmp.h>
 #include <sys/types.h>
+#include <assert.h>
 
 #ifdef HAVE_SYS_TIME_H
 #include <sys/time.h>
@@ -35,7 +37,12 @@
 
 #if defined _WIN32 || defined __CYGWIN__
 #include <windows.h>
+#elif defined(HAVE_POSIX_MEMALIGN)
+#elif defined(HAVE_MEMALIGN)
+#include <malloc.h>
 #endif
+static void aligned_free(void *);
+static void *aligned_malloc(size_t alignment, size_t size);
 
 #ifdef HAVE_VALGRIND_MEMCHECK_H
 # include <valgrind/memcheck.h>
@@ -321,6 +328,24 @@ struct gc_list {
 
 #define CALC_EXACT_MALLOC_SIZE 0
 
+#ifdef POOL_ALLOC_API
+/* POOL ALLOC API */
+#define POOL_ALLOC_PART 1
+#include "pool_alloc.inc.h"
+#undef POOL_ALLOC_PART
+
+typedef struct pool_layout_t pool_layout_t;
+struct pool_layout_t {
+    pool_header
+      p6,  /* st_table && st_table_entry */
+      p11;  /* st_table.bins init size */
+} pool_layout = {
+    INIT_POOL(void*[6]),
+    INIT_POOL(void*[11])
+};
+static void pool_finalize_header(pool_header *header);
+#endif
+
 typedef struct rb_objspace {
     struct {
 	size_t limit;
@@ -330,6 +355,9 @@ typedef struct rb_objspace {
 	size_t allocations;
 #endif
     } malloc_params;
+#ifdef POOL_ALLOC_API
+    pool_layout_t *pool_headers;
+#endif
     struct {
 	size_t increment;
 	struct heaps_slot *ptr;
@@ -377,7 +405,11 @@ typedef struct rb_objspace {
 #define ruby_initial_gc_stress	initial_params.gc_stress
 int *ruby_initial_gc_stress_ptr = &ruby_initial_gc_stress;
 #else
+#  ifdef POOL_ALLOC_API
+static rb_objspace_t rb_objspace = {{GC_MALLOC_LIMIT}, &pool_layout, {HEAP_MIN_SLOTS}};
+#  else
 static rb_objspace_t rb_objspace = {{GC_MALLOC_LIMIT}, {HEAP_MIN_SLOTS}};
+#  endif
 int *ruby_initial_gc_stress_ptr = &rb_objspace.gc_stress;
 #endif
 #define malloc_limit		objspace->malloc_params.limit
@@ -413,6 +445,10 @@ rb_objspace_alloc(void)
     memset(objspace, 0, sizeof(*objspace));
     malloc_limit = initial_malloc_limit;
     ruby_gc_stress = ruby_initial_gc_stress;
+#ifdef POOL_ALLOC_API
+    objspace->pool_headers = (pool_layout_t*) malloc(sizeof(pool_layout));
+    memcpy(objspace->pool_headers, &pool_layout, sizeof(pool_layout));
+#endif
 
     return objspace;
 }
@@ -491,6 +527,13 @@ rb_objspace_free(rb_objspace_t *objspace)
 	heaps_used = 0;
 	heaps = 0;
     }
+#ifdef POOL_ALLOC_API
+    if (objspace->pool_headers) {
+        pool_finalize_header(&objspace->pool_headers->p6);
+        pool_finalize_header(&objspace->pool_headers->p11);
+        free(objspace->pool_headers);
+    }
+#endif
     free(objspace);
 }
 #endif
@@ -894,6 +937,27 @@ ruby_xfree(void *x)
 	vm_xfree(&rb_objspace, x);
 }
 
+#ifdef POOL_ALLOC_API
+/* POOL ALLOC API */
+#define POOL_ALLOC_PART 2
+#include "pool_alloc.inc.h"
+#undef POOL_ALLOC_PART
+
+void
+ruby_xpool_free(void *ptr)
+{
+    pool_free_entry((void**)ptr);
+}
+
+#define CONCRET_POOL_MALLOC(pnts) \
+void * ruby_xpool_malloc_##pnts##p () { \
+    return pool_alloc_entry(&rb_objspace.pool_headers->p##pnts ); \
+}
+CONCRET_POOL_MALLOC(6)
+CONCRET_POOL_MALLOC(11)
+#undef CONCRET_POOL_MALLOC
+
+#endif
 
 /*
  *  call-seq:
@@ -1008,6 +1072,55 @@ allocate_sorted_heaps(rb_objspace_t *objspace, size_t next_heaps_length)
     heaps_length = next_heaps_length;
 }
 
+static void *
+aligned_malloc(size_t alignment, size_t size)
+{
+    void *res;
+
+#if defined __MINGW32__
+    res = __mingw_aligned_malloc(size, alignment);
+#elif defined _WIN32 && !defined __CYGWIN__
+    res = _aligned_malloc(size, alignment);
+#elif defined(HAVE_POSIX_MEMALIGN)
+    if (posix_memalign(&res, alignment, size) == 0) {
+        return res;
+    }
+    else {
+        return NULL;
+    }
+#elif defined(HAVE_MEMALIGN)
+    res = memalign(alignment, size);
+#else
+    char* aligned;
+    res = malloc(alignment + size + sizeof(void*));
+    aligned = (char*)res + alignment + sizeof(void*);
+    aligned -= ((VALUE)aligned & (alignment - 1));
+    ((void**)aligned)[-1] = res;
+    res = (void*)aligned;
+#endif
+
+#if defined(_DEBUG) || defined(GC_DEBUG)
+    /* alignment must be a power of 2 */
+    assert((alignment - 1) & alignment == 0);
+    assert(alignment % sizeof(void*) == 0);
+#endif
+    return res;
+}
+
+static void
+aligned_free(void *ptr)
+{
+#if defined __MINGW32__
+    __mingw_aligned_free(ptr);
+#elif defined _WIN32 && !defined __CYGWIN__
+    _aligned_free(ptr);
+#elif defined(HAVE_MEMALIGN) || defined(HAVE_POSIX_MEMALIGN)
+    free(ptr);
+#else
+    free(((void**)ptr)[-1]);
+#endif
+}
+
 static void
 assign_heap_slot(rb_objspace_t *objspace)
 {
diff --git a/hash.c b/hash.c
index fbd8237..32917c3 100644
--- a/hash.c
+++ b/hash.c
@@ -44,7 +44,7 @@ rb_any_cmp(VALUE a, VALUE b)
     if (FIXNUM_P(a) && FIXNUM_P(b)) {
 	return a != b;
     }
-    if (TYPE(a) == T_STRING && RBASIC(a)->klass == rb_cString &&
+    if (RB_TYPE_P(a, T_STRING) && RBASIC(a)->klass == rb_cString &&
 	TYPE(b) == T_STRING && RBASIC(b)->klass == rb_cString) {
 	return rb_str_hash_cmp(a, b);
     }
@@ -80,20 +80,14 @@ rb_any_hash(VALUE a)
     VALUE hval;
     st_index_t hnum;
 
-    switch (TYPE(a)) {
-      case T_FIXNUM:
-      case T_SYMBOL:
-      case T_NIL:
-      case T_FALSE:
-      case T_TRUE:
-	hnum = rb_hash_end(rb_hash_start((unsigned int)a));
-	break;
-
-      case T_STRING:
+    if (SPECIAL_CONST_P(a)) {
+	if (a == Qundef) return 0;
+	hnum = rb_hash_end(rb_hash_start((st_index_t)a));
+    }
+    else if (BUILTIN_TYPE(a) == T_STRING) {
 	hnum = rb_str_hash(a);
-	break;
-
-      default:
+    }
+    else {
         hval = rb_hash(a);
 	hnum = FIX2LONG(hval);
     }
@@ -106,10 +100,8 @@ static const struct st_hash_type objhash = {
     rb_any_hash,
 };
 
-static const struct st_hash_type identhash = {
-    st_numcmp,
-    st_numhash,
-};
+extern const struct st_hash_type st_hashtype_num;
+#define identhash st_hashtype_num
 
 typedef int st_foreach_func(st_data_t, st_data_t, st_data_t);
 
@@ -124,7 +116,6 @@ foreach_safe_i(st_data_t key, st_data_t value, struct foreach_safe_arg *arg)
 {
     int status;
 
-    if (key == Qundef) return ST_CONTINUE;
     status = (*arg->func)(key, value, arg->arg);
     if (status == ST_CONTINUE) {
 	return ST_CHECK;
@@ -140,7 +131,7 @@ st_foreach_safe(st_table *table, int (*func)(ANYARGS), st_data_t a)
     arg.tbl = table;
     arg.func = (st_foreach_func *)func;
     arg.arg = a;
-    if (st_foreach(table, foreach_safe_i, (st_data_t)&arg)) {
+    if (st_foreach_check(table, foreach_safe_i, (st_data_t)&arg, 0)) {
 	rb_raise(rb_eRuntimeError, "hash modified during iteration");
     }
 }
@@ -154,21 +145,21 @@ struct hash_foreach_arg {
 };
 
 static int
-hash_foreach_iter(st_data_t key, st_data_t value, struct hash_foreach_arg *arg)
+hash_foreach_iter(st_data_t key, st_data_t value, st_data_t argp)
 {
+    struct hash_foreach_arg *arg = (struct hash_foreach_arg *)argp;
     int status;
     st_table *tbl;
 
     tbl = RHASH(arg->hash)->ntbl;
-    if ((VALUE)key == Qundef) return ST_CONTINUE;
     status = (*arg->func)((VALUE)key, (VALUE)value, arg->arg);
     if (RHASH(arg->hash)->ntbl != tbl) {
 	rb_raise(rb_eRuntimeError, "rehash occurred during iteration");
     }
     switch (status) {
       case ST_DELETE:
-	st_delete_safe(tbl, &key, 0, Qundef);
 	FL_SET(arg->hash, HASH_DELETED);
+	return ST_DELETE;
       case ST_CONTINUE:
 	break;
       case ST_STOP:
@@ -184,7 +175,7 @@ hash_foreach_ensure(VALUE hash)
 
     if (RHASH(hash)->iter_lev == 0) {
 	if (FL_TEST(hash, HASH_DELETED)) {
-	    st_cleanup_safe(RHASH(hash)->ntbl, Qundef);
+	    st_cleanup_safe(RHASH(hash)->ntbl, (st_data_t)Qundef);
 	    FL_UNSET(hash, HASH_DELETED);
 	}
     }
@@ -192,9 +183,10 @@ hash_foreach_ensure(VALUE hash)
 }
 
 static VALUE
-hash_foreach_call(struct hash_foreach_arg *arg)
+hash_foreach_call(VALUE arg)
 {
-    if (st_foreach(RHASH(arg->hash)->ntbl, hash_foreach_iter, (st_data_t)arg)) {
+    VALUE hash = ((struct hash_foreach_arg *)arg)->hash;
+    if (st_foreach_check(RHASH(hash)->ntbl, hash_foreach_iter, (st_data_t)arg, (st_data_t)Qundef)) {
 	rb_raise(rb_eRuntimeError, "hash modified during iteration");
     }
     return Qnil;
@@ -447,7 +439,7 @@ rb_hash_rehash_i(VALUE key, VALUE value, VALUE arg)
 {
     st_table *tbl = (st_table *)arg;
 
-    if (key != Qundef) st_insert(tbl, key, value);
+    st_insert(tbl, (st_data_t)key, (st_data_t)value);
     return ST_CONTINUE;
 }
 
@@ -490,6 +482,20 @@ rb_hash_rehash(VALUE hash)
     return hash;
 }
 
+static VALUE
+hash_default_value(VALUE hash, VALUE key)
+{
+    if (rb_method_basic_definition_p(CLASS_OF(hash), id_default)) {
+	VALUE ifnone = RHASH_IFNONE(hash);
+	if (!FL_TEST(hash, HASH_PROC_DEFAULT)) return ifnone;
+	if (key == Qundef) return Qnil;
+	return rb_funcall(ifnone, id_yield, 2, hash, key);
+    }
+    else {
+	return rb_funcall(hash, id_default, 1, key);
+    }
+}
+
 /*
  *  call-seq:
  *     hsh[key]    ->  value
@@ -510,13 +516,7 @@ rb_hash_aref(VALUE hash, VALUE key)
     st_data_t val;
 
     if (!RHASH(hash)->ntbl || !st_lookup(RHASH(hash)->ntbl, key, &val)) {
-	if (!FL_TEST(hash, HASH_PROC_DEFAULT) &&
-	    rb_method_basic_definition_p(CLASS_OF(hash), id_default)) {
-	    return RHASH_IFNONE(hash);
-	}
-	else {
-	    return rb_funcall(hash, id_default, 1, key);
-	}
+	return hash_default_value(hash, key);
     }
     return (VALUE)val;
 }
@@ -659,7 +659,7 @@ rb_hash_default(int argc, VALUE *argv, VALUE hash)
 static VALUE
 rb_hash_set_default(VALUE hash, VALUE ifnone)
 {
-    rb_hash_modify(hash);
+    rb_hash_modify_check(hash);
     RHASH_IFNONE(hash) = ifnone;
     FL_UNSET(hash, HASH_PROC_DEFAULT);
     return ifnone;
@@ -707,7 +707,7 @@ rb_hash_set_default_proc(VALUE hash, VALUE proc)
 {
     VALUE b;
 
-    rb_hash_modify(hash);
+    rb_hash_modify_check(hash);
     b = rb_check_convert_type(proc, T_DATA, "Proc", "to_proc");
     if (NIL_P(b) || !rb_obj_is_proc(b)) {
 	rb_raise(rb_eTypeError,
@@ -776,7 +776,7 @@ rb_hash_delete_key(VALUE hash, VALUE key)
     if (!RHASH(hash)->ntbl)
         return Qundef;
     if (RHASH(hash)->iter_lev > 0) {
-	if (st_delete_safe(RHASH(hash)->ntbl, &ktmp, &val, Qundef)) {
+	if (st_delete_safe(RHASH(hash)->ntbl, &ktmp, &val, (st_data_t)Qundef)) {
 	    FL_SET(hash, HASH_DELETED);
 	    return (VALUE)val;
 	}
@@ -809,7 +809,7 @@ rb_hash_delete(VALUE hash, VALUE key)
 {
     VALUE val;
 
-    rb_hash_modify(hash);
+    rb_hash_modify_check(hash);
     val = rb_hash_delete_key(hash, key);
     if (val != Qundef) return val;
     if (rb_block_given_p()) {
@@ -828,7 +828,6 @@ shift_i(VALUE key, VALUE value, VALUE arg)
 {
     struct shift_var *var = (struct shift_var *)arg;
 
-    if (key == Qundef) return ST_CONTINUE;
     if (var->key != Qundef) return ST_STOP;
     var->key = key;
     var->val = value;
@@ -840,7 +839,6 @@ shift_i_safe(VALUE key, VALUE value, VALUE arg)
 {
     struct shift_var *var = (struct shift_var *)arg;
 
-    if (key == Qundef) return ST_CONTINUE;
     var->key = key;
     var->val = value;
     return ST_STOP;
@@ -864,29 +862,25 @@ rb_hash_shift(VALUE hash)
 {
     struct shift_var var;
 
-    rb_hash_modify(hash);
-    var.key = Qundef;
-    rb_hash_foreach(hash, RHASH(hash)->iter_lev > 0 ? shift_i_safe : shift_i,
-		    (VALUE)&var);
-
-    if (var.key != Qundef) {
-	if (RHASH(hash)->iter_lev > 0) {
-	    rb_hash_delete_key(hash, var.key);
+    rb_hash_modify_check(hash);
+    if (RHASH(hash)->ntbl) {
+	var.key = Qundef;
+	rb_hash_foreach(hash, RHASH(hash)->iter_lev > 0 ? shift_i_safe : shift_i,
+			(VALUE)&var);
+
+	if (var.key != Qundef) {
+	    if (RHASH(hash)->iter_lev > 0) {
+		rb_hash_delete_key(hash, var.key);
+	    }
+	    return rb_assoc_new(var.key, var.val);
 	}
-	return rb_assoc_new(var.key, var.val);
-    }
-    else if (FL_TEST(hash, HASH_PROC_DEFAULT)) {
-	return rb_funcall(RHASH_IFNONE(hash), id_yield, 2, hash, Qnil);
-    }
-    else {
-	return RHASH_IFNONE(hash);
     }
+    return hash_default_value(hash, Qnil);
 }
 
 static int
 delete_if_i(VALUE key, VALUE value, VALUE hash)
 {
-    if (key == Qundef) return ST_CONTINUE;
     if (RTEST(rb_yield_values(2, key, value))) {
 	rb_hash_delete_key(hash, key);
     }
@@ -912,8 +906,9 @@ VALUE
 rb_hash_delete_if(VALUE hash)
 {
     RETURN_ENUMERATOR(hash, 0, 0);
-    rb_hash_modify(hash);
-    rb_hash_foreach(hash, delete_if_i, hash);
+    rb_hash_modify_check(hash);
+    if (RHASH(hash)->ntbl)
+	rb_hash_foreach(hash, delete_if_i, hash);
     return hash;
 }
 
@@ -984,7 +979,6 @@ rb_hash_values_at(int argc, VALUE *argv, VALUE hash)
 static int
 select_i(VALUE key, VALUE value, VALUE result)
 {
-    if (key == Qundef) return ST_CONTINUE;
     if (RTEST(rb_yield_values(2, key, value)))
 	rb_hash_aset(result, key, value);
     return ST_CONTINUE;
@@ -1018,7 +1012,6 @@ rb_hash_select(VALUE hash)
 static int
 keep_if_i(VALUE key, VALUE value, VALUE hash)
 {
-    if (key == Qundef) return ST_CONTINUE;
     if (!RTEST(rb_yield_values(2, key, value))) {
 	return ST_DELETE;
     }
@@ -1040,7 +1033,7 @@ rb_hash_select_bang(VALUE hash)
     st_index_t n;
 
     RETURN_ENUMERATOR(hash, 0, 0);
-    rb_hash_modify(hash);
+    rb_hash_modify_check(hash);
     if (!RHASH(hash)->ntbl)
         return Qnil;
     n = RHASH(hash)->ntbl->num_entries;
@@ -1065,8 +1058,9 @@ VALUE
 rb_hash_keep_if(VALUE hash)
 {
     RETURN_ENUMERATOR(hash, 0, 0);
-    rb_hash_modify(hash);
-    rb_hash_foreach(hash, keep_if_i, hash);
+    rb_hash_modify_check(hash);
+    if (RHASH(hash)->ntbl)
+	rb_hash_foreach(hash, keep_if_i, hash);
     return hash;
 }
 
@@ -1144,9 +1138,7 @@ rb_hash_aset(VALUE hash, VALUE key, VALUE val)
 static int
 replace_i(VALUE key, VALUE val, VALUE hash)
 {
-    if (key != Qundef) {
-	rb_hash_aset(hash, key, val);
-    }
+    rb_hash_aset(hash, key, val);
 
     return ST_CONTINUE;
 }
@@ -1227,7 +1219,6 @@ rb_hash_empty_p(VALUE hash)
 static int
 each_value_i(VALUE key, VALUE value)
 {
-    if (key == Qundef) return ST_CONTINUE;
     rb_yield(value);
     return ST_CONTINUE;
 }
@@ -1262,7 +1253,6 @@ rb_hash_each_value(VALUE hash)
 static int
 each_key_i(VALUE key, VALUE value)
 {
-    if (key == Qundef) return ST_CONTINUE;
     rb_yield(key);
     return ST_CONTINUE;
 }
@@ -1296,7 +1286,6 @@ rb_hash_each_key(VALUE hash)
 static int
 each_pair_i(VALUE key, VALUE value)
 {
-    if (key == Qundef) return ST_CONTINUE;
     rb_yield(rb_assoc_new(key, value));
     return ST_CONTINUE;
 }
@@ -1334,7 +1323,6 @@ rb_hash_each_pair(VALUE hash)
 static int
 to_a_i(VALUE key, VALUE value, VALUE ary)
 {
-    if (key == Qundef) return ST_CONTINUE;
     rb_ary_push(ary, rb_assoc_new(key, value));
     return ST_CONTINUE;
 }
@@ -1367,7 +1355,6 @@ inspect_i(VALUE key, VALUE value, VALUE str)
 {
     VALUE str2;
 
-    if (key == Qundef) return ST_CONTINUE;
     str2 = rb_inspect(key);
     if (RSTRING_LEN(str) > 1) {
 	rb_str_cat2(str, ", ");
@@ -1434,7 +1421,6 @@ rb_hash_to_hash(VALUE hash)
 static int
 keys_i(VALUE key, VALUE value, VALUE ary)
 {
-    if (key == Qundef) return ST_CONTINUE;
     rb_ary_push(ary, key);
     return ST_CONTINUE;
 }
@@ -1465,7 +1451,6 @@ rb_hash_keys(VALUE hash)
 static int
 values_i(VALUE key, VALUE value, VALUE ary)
 {
-    if (key == Qundef) return ST_CONTINUE;
     rb_ary_push(ary, value);
     return ST_CONTINUE;
 }
@@ -1524,7 +1509,6 @@ rb_hash_search_value(VALUE key, VALUE value, VALUE arg)
 {
     VALUE *data = (VALUE *)arg;
 
-    if (key == Qundef) return ST_CONTINUE;
     if (rb_equal(value, data[1])) {
 	data[0] = Qtrue;
 	return ST_STOP;
@@ -1568,7 +1552,6 @@ eql_i(VALUE key, VALUE val1, VALUE arg)
     struct equal_data *data = (struct equal_data *)arg;
     st_data_t val2;
 
-    if (key == Qundef) return ST_CONTINUE;
     if (!st_lookup(data->tbl, key, &val2)) {
 	data->result = Qfalse;
 	return ST_STOP;
@@ -1599,7 +1582,7 @@ hash_equal(VALUE hash1, VALUE hash2, int eql)
     struct equal_data data;
 
     if (hash1 == hash2) return Qtrue;
-    if (TYPE(hash2) != T_HASH) {
+    if (!RB_TYPE_P(hash2, T_HASH)) {
 	if (!rb_respond_to(hash2, rb_intern("to_hash"))) {
 	    return Qfalse;
 	}
@@ -1670,7 +1653,6 @@ hash_i(VALUE key, VALUE val, VALUE arg)
     st_index_t *hval = (st_index_t *)arg;
     st_index_t hdata[2];
 
-    if (key == Qundef) return ST_CONTINUE;
     hdata[0] = rb_hash(key);
     hdata[1] = rb_hash(val);
     *hval ^= st_hash(hdata, sizeof(hdata), 0);
@@ -1711,7 +1693,6 @@ rb_hash_hash(VALUE hash)
 static int
 rb_hash_invert_i(VALUE key, VALUE value, VALUE hash)
 {
-    if (key == Qundef) return ST_CONTINUE;
     rb_hash_aset(hash, value, key);
     return ST_CONTINUE;
 }
@@ -1740,7 +1721,6 @@ rb_hash_invert(VALUE hash)
 static int
 rb_hash_update_i(VALUE key, VALUE value, VALUE hash)
 {
-    if (key == Qundef) return ST_CONTINUE;
     hash_update(hash, key);
     st_insert(RHASH(hash)->ntbl, key, value);
     return ST_CONTINUE;
@@ -1749,7 +1729,6 @@ rb_hash_update_i(VALUE key, VALUE value, VALUE hash)
 static int
 rb_hash_update_block_i(VALUE key, VALUE value, VALUE hash)
 {
-    if (key == Qundef) return ST_CONTINUE;
     if (rb_hash_has_key(hash, key)) {
 	value = rb_yield_values(3, key, rb_hash_aref(hash, key), value);
     }
@@ -1806,7 +1785,6 @@ rb_hash_update_func_i(VALUE key, VALUE value, VALUE arg0)
     struct update_arg *arg = (struct update_arg *)arg0;
     VALUE hash = arg->hash;
 
-    if (key == Qundef) return ST_CONTINUE;
     if (rb_hash_has_key(hash, key)) {
 	value = (*arg->func)(key, rb_hash_aref(hash, key), value);
     }
@@ -1863,7 +1841,6 @@ assoc_i(VALUE key, VALUE val, VALUE arg)
 {
     VALUE *args = (VALUE *)arg;
 
-    if (key == Qundef) return ST_CONTINUE;
     if (RTEST(rb_equal(args[0], key))) {
 	args[1] = rb_assoc_new(key, val);
 	return ST_STOP;
@@ -1901,7 +1878,6 @@ rassoc_i(VALUE key, VALUE val, VALUE arg)
 {
     VALUE *args = (VALUE *)arg;
 
-    if (key == Qundef) return ST_CONTINUE;
     if (RTEST(rb_equal(args[0], val))) {
 	args[1] = rb_assoc_new(key, val);
 	return ST_STOP;
@@ -2198,7 +2174,7 @@ rb_env_path_tainted(void)
 }
 
 #if defined(_WIN32) || (defined(HAVE_SETENV) && defined(HAVE_UNSETENV))
-#elif defined __sun__
+#elif defined __sun
 static int
 in_origenv(const char *str)
 {
@@ -2286,7 +2262,7 @@ ruby_setenv(const char *name, const char *value)
 	    rb_sys_fail("unsetenv");
 #endif
     }
-#elif defined __sun__
+#elif defined __sun
     size_t len;
     char **env_ptr, *str;
     if (strchr(name, '=')) {
@@ -3084,11 +3060,9 @@ env_invert(void)
 static int
 env_replace_i(VALUE key, VALUE val, VALUE keys)
 {
-    if (key != Qundef) {
-	env_aset(Qnil, key, val);
-	if (rb_ary_includes(keys, key)) {
-	    rb_ary_delete(keys, key);
-	}
+    env_aset(Qnil, key, val);
+    if (rb_ary_includes(keys, key)) {
+	rb_ary_delete(keys, key);
     }
     return ST_CONTINUE;
 }
@@ -3120,12 +3094,10 @@ env_replace(VALUE env, VALUE hash)
 static int
 env_update_i(VALUE key, VALUE val)
 {
-    if (key != Qundef) {
-	if (rb_block_given_p()) {
-	    val = rb_yield_values(3, key, rb_f_getenv(Qnil, key), val);
-	}
-	env_aset(Qnil, key, val);
+    if (rb_block_given_p()) {
+	val = rb_yield_values(3, key, rb_f_getenv(Qnil, key), val);
     }
+    env_aset(Qnil, key, val);
     return ST_CONTINUE;
 }
 
@@ -3150,15 +3122,116 @@ env_update(VALUE env, VALUE hash)
 }
 
 /*
- *  A <code>Hash</code> is a collection of key-value pairs. It is
- *  similar to an <code>Array</code>, except that indexing is done via
- *  arbitrary keys of any object type, not an integer index. Hashes enumerate
- *  their values in the order that the corresponding keys were inserted.
+ *  A Hash is a dictionary-like collection of unique keys and their values.
+ *  Also called associative arrays, they are similar to Arrays, but where an
+ *  Array uses integers as its index, a Hash allows you to use any object
+ *  type.
+ *
+ *  Hashes enumerate their values in the order that the corresponding keys
+ *  were inserted.
+ *
+ *  A Hash can be easily created by using its implicit form:
+ *
+ *    grades = { "Jane Doe" => 10, "Jim Doe" => 6 }
+ *
+ *  Hashes allow an alternate syntax form when your keys are always symbols.
+ *  Instead of
+ *
+ *    options = { :font_size => 10, :font_family => "Arial" }
+ *
+ *  You could write it as:
+ *
+ *    options = { font_size: 10, font_family: "Arial" }
+ *
+ *  Each named key is a symbol you can access in hash:
+ *
+ *    options[:font_size]  # => 10
+ *
+ *  A Hash can also be created through its ::new method:
+ *
+ *    grades = Hash.new
+ *    grades["Dorothy Doe"] = 9
  *
  *  Hashes have a <em>default value</em> that is returned when accessing
- *  keys that do not exist in the hash. By default, that value is
- *  <code>nil</code>.
+ *  keys that do not exist in the hash. If no default is set +nil+ is used.
+ *  You can set the default value by sending it as an argument to Hash.new:
+ *
+ *    grades = Hash.new(0)
+ *
+ *  Or by using the #default= method:
+ *
+ *    grades = {"Timmy Doe" => 8}
+ *    grades.default = 0
+ *
+ *  Accessing a value in a Hash requires using its key:
+ *
+ *    puts grades["Jane Doe"] # => 10
+ *
+ *  === Common Uses
+ *
+ *  Hashes are an easy way to represent data structures, such as
+ *
+ *    books         = {}
+ *    books[:matz]  = "The Ruby Language"
+ *    books[:black] = "The Well-Grounded Rubyist"
+ *
+ *  Hashes are also commonly used as a way to have named parameters in
+ *  functions. Note that no brackets are used below. If a hash is the last
+ *  argument on a method call, no braces are needed, thus creating a really
+ *  clean interface:
+ *
+ *    Person.create(name: "John Doe", age: 27)
+ *
+ *    def self.create(params)
+ *      @name = params[:name]
+ *      @age  = params[:age]
+ *    end
+ *
+ *  === Hash Keys
+ *
+ *  Two objects refer to the same hash key when their <code>hash</code> value
+ *  is identical and the two objects are <code>eql?</code> to each other.
+ *
+ *  A user-defined class may be used as a hash key if the <code>hash</code>
+ *  and <code>eql?</code> methods are overridden to provide meaningful
+ *  behavior.  By default, separate instances refer to separate hash keys.
+ *
+ *  A typical implementation of <code>hash</code> is based on the
+ *  object's data while <code>eql?</code> is usually aliased to the overridden
+ *  <code>==</code> method:
+ *
+ *    class Book
+ *      attr_reader :author, :title
+ *
+ *      def initialize(author, title)
+ *        @author = author
+ *        @title = title
+ *      end
+ *
+ *      def ==(other)
+ *        self.class === other and
+ *          other.author == @author and
+ *          other.title == @title
+ *      end
+ *
+ *      alias eql? ==
+ *
+ *      def hash
+ *        @author.hash ^ @title.hash # XOR
+ *      end
+ *    end
+ *
+ *    book1 = Book.new 'matz', 'Ruby in a Nutshell'
+ *    book2 = Book.new 'matz', 'Ruby in a Nutshell'
+ *
+ *    reviews = {}
+ *
+ *    reviews[book1] = 'Great reference!'
+ *    reviews[book2] = 'Nice and compact!'
+ *
+ *    reviews.length #=> 1
  *
+ *  See also Object#hash and Object#eql?
  */
 
 void
diff --git a/include/ruby/st.h b/include/ruby/st.h
index 50f2a75..119dfde 100644
--- a/include/ruby/st.h
+++ b/include/ruby/st.h
@@ -36,7 +36,7 @@ typedef unsigned long st_data_t;
 #elif SIZEOF_LONG_LONG == SIZEOF_VOIDP
 typedef unsigned LONG_LONG st_data_t;
 #else
-# error ---->> st.c requires sizeof(void*) == sizeof(long) to be compiled. <<----
+# error ---->> st.c requires sizeof(void*) == sizeof(long) or sizeof(LONG_LONG) to be compiled. <<----
 #endif
 #define ST_DATA_T_DEFINED
 
@@ -74,6 +74,11 @@ struct st_hash_type {
 
 #define ST_INDEX_BITS (sizeof(st_index_t) * CHAR_BIT)
 
+typedef struct st_packed_entry {
+    st_index_t hash;
+    st_data_t key, val;
+} st_packed_entry;
+
 struct st_table {
     const struct st_hash_type *type;
     st_index_t num_bins;
@@ -91,8 +96,17 @@ struct st_table {
     __extension__
 #endif
     st_index_t num_entries : ST_INDEX_BITS - 1;
-    struct st_table_entry **bins;
-    struct st_table_entry *head, *tail;
+    union {
+	struct {
+	    struct st_table_entry **bins;
+	    struct st_table_entry *head, *tail;
+	} big;
+	struct {
+	    struct st_packed_entry *entries;
+	    st_index_t real_entries;
+	} packed;
+	st_packed_entry upacked;
+    } as;
 };
 
 #define st_is_member(table,key) st_lookup((table),(key),(st_data_t *)0)
@@ -114,6 +128,7 @@ int st_insert2(st_table *, st_data_t, st_data_t, st_data_t (*)(st_data_t));
 int st_lookup(st_table *, st_data_t, st_data_t *);
 int st_get_key(st_table *, st_data_t, st_data_t *);
 int st_foreach(st_table *, int (*)(ANYARGS), st_data_t);
+int st_foreach_check(st_table *, int (*)(ANYARGS), st_data_t, st_data_t);
 int st_reverse_foreach(st_table *, int (*)(ANYARGS), st_data_t);
 void st_add_direct(st_table *, st_data_t, st_data_t);
 void st_free_table(st_table *);
diff --git a/pool_alloc.h b/pool_alloc.h
new file mode 100644
index 0000000..957708e
--- /dev/null
+++ b/pool_alloc.h
@@ -0,0 +1,11 @@
+#ifndef POOL_ALLOC_H
+#define POOL_ALLOC_H
+
+#define POOL_ALLOC_API
+#ifdef POOL_ALLOC_API
+void  ruby_xpool_free(void *ptr);
+void *ruby_xpool_malloc_6p();
+void *ruby_xpool_malloc_11p();
+#endif
+
+#endif
diff --git a/pool_alloc.inc.h b/pool_alloc.inc.h
new file mode 100644
index 0000000..a7879ab
--- /dev/null
+++ b/pool_alloc.inc.h
@@ -0,0 +1,156 @@
+/*
+ * this is generic pool allocator
+ * you should define following macroses:
+ * ITEM_NAME - unique identifier, which allows to hold functions in a namespace
+ * ITEM_TYPEDEF(name) - passed to typedef to localize item type
+ * free_entry - desired name of function for free entry
+ * alloc_entry - defired name of function for allocate entry
+ */
+
+#if POOL_ALLOC_PART == 1
+#ifdef HEAP_ALIGN_LOG
+#define DEFAULT_POOL_SIZE (1 << HEAP_ALIGN_LOG)
+#else
+#define DEFAULT_POOL_SIZE (sizeof(void*) * 2048)
+#endif
+typedef unsigned int pool_holder_counter;
+
+typedef struct pool_entry_list pool_entry_list;
+typedef struct pool_holder pool_holder;
+
+typedef struct pool_header {
+    pool_holder         *first;
+    pool_holder         *_black_magick;
+    pool_holder_counter  size; // size of entry in sizeof(void*) items
+    pool_holder_counter  total; // size of entry in sizeof(void*) items
+} pool_header;
+
+struct pool_holder {
+    pool_holder_counter free, total;
+    pool_header  *header;
+    void               *freep;
+    pool_holder        *fore, *back;
+    void *data[1];
+};
+#define POOL_DATA_SIZE(pool_size) (((pool_size) - sizeof(void*) * 6 - offsetof(pool_holder, data)) / sizeof(void*))
+#define POOL_ENTRY_SIZE(item_type) ((sizeof(item_type) - 1) / sizeof(void*) + 1)
+#define POOL_HOLDER_COUNT(pool_size, item_type) (POOL_DATA_SIZE(pool_size)/POOL_ENTRY_SIZE(item_type))
+#define INIT_POOL(item_type) {NULL, NULL, POOL_ENTRY_SIZE(item_type), POOL_HOLDER_COUNT(DEFAULT_POOL_SIZE, item_type)}
+
+#elif POOL_ALLOC_PART == 2
+static pool_holder *
+pool_holder_alloc(pool_header *header)
+{
+    pool_holder *holder;
+    pool_holder_counter i, size, count;
+    register void **ptr;
+
+    size_t sz = offsetof(pool_holder, data) +
+	    header->size * header->total * sizeof(void*);
+#define objspace (&rb_objspace)
+    vm_malloc_prepare(objspace, DEFAULT_POOL_SIZE);
+    if (header->first != NULL) return header->first;
+    TRY_WITH_GC(holder = (pool_holder*) aligned_malloc(DEFAULT_POOL_SIZE, sz));
+    malloc_increase += DEFAULT_POOL_SIZE;
+#if CALC_EXACT_MALLOC_SIZE
+    objspace->malloc_params.allocated_size += DEFAULT_POOL_SIZE;
+    objspace->malloc_params.allocations++;
+#endif
+#undef objspace
+
+    size = header->size;
+    count = header->total;
+    holder->free = count;
+    holder->total = count;
+    holder->header = header;
+    holder->fore = NULL;
+    holder->back = NULL;
+    holder->freep = &holder->data;
+    ptr = holder->data;
+    for(i = count - 1; i; i-- ) {
+	ptr = *ptr = ptr + size;
+    }
+    *ptr = NULL;
+    header->first = holder;
+    return holder;
+}
+
+static inline void
+pool_holder_unchaing(pool_header *header, pool_holder *holder)
+{
+    register pool_holder *fore = holder->fore, *back = holder->back;
+    holder->fore = NULL;
+    holder->back = NULL;
+    if (fore != NULL)  fore->back     = back;
+    else               header->_black_magick = back;
+    if (back != NULL)  back->fore     = fore;
+    else               header->first = fore;
+}
+
+static inline pool_holder *
+entry_holder(void **entry)
+{
+    return (pool_holder*)(((uintptr_t)entry) & ~(DEFAULT_POOL_SIZE - 1));
+}
+
+static inline void
+pool_free_entry(void **entry)
+{
+    pool_holder *holder = entry_holder(entry);
+    pool_header *header = holder->header;
+
+    if (holder->free++ == 0) {
+	register pool_holder *first = header->first;
+	if (first == NULL) {
+	    header->first = holder;
+	} else {
+	    holder->back = first;
+	    holder->fore = first->fore;
+	    first->fore = holder;
+	    if (holder->fore)
+		holder->fore->back = holder;
+	    else
+		header->_black_magick = holder;
+	}
+    } else if (holder->free == holder->total && header->first != holder ) {
+	pool_holder_unchaing(header, holder);
+	aligned_free(holder);
+#if CALC_EXACT_MALLOC_SIZE
+	rb_objspace.malloc_params.allocated_size -= DEFAULT_POOL_SIZE;
+	rb_objspace.malloc_params.allocations--;
+#endif
+	return;
+    }
+
+    *entry = holder->freep;
+    holder->freep = entry;
+}
+
+static inline void*
+pool_alloc_entry(pool_header *header)
+{
+    pool_holder *holder = header->first;
+    void **result;
+    if (holder == NULL) {
+	holder = pool_holder_alloc(header);
+    }
+
+    result = holder->freep;
+    holder->freep = *result;
+
+    if (--holder->free == 0) {
+	pool_holder_unchaing(header, holder);
+    }
+
+    return result;
+}
+
+static void
+pool_finalize_header(pool_header *header)
+{
+    if (header->first) {
+        aligned_free(header->first);
+        header->first = NULL;
+    }
+}
+#endif
diff --git a/st.c b/st.c
index fda5784..20ec427 100644
--- a/st.c
+++ b/st.c
@@ -7,6 +7,7 @@
 #include "st.h"
 #else
 #include "ruby/ruby.h"
+#include "pool_alloc.h"
 #endif
 
 #include <stdio.h>
@@ -25,8 +26,17 @@ struct st_table_entry {
     st_table_entry *fore, *back;
 };
 
-#define ST_DEFAULT_MAX_DENSITY 5
+#define STATIC_ASSERT(name, expr) typedef int static_assert_##name##_check[(expr) ? 1 : -1];
+
+#define ST_DEFAULT_MAX_DENSITY 2
 #define ST_DEFAULT_INIT_TABLE_SIZE 11
+#define ST_DEFAULT_SECOND_TABLE_SIZE 19
+#define ST_DEFAULT_PACKED_TABLE_SIZE 19
+#define PACKED_UNIT (int)(sizeof(st_packed_entry) / sizeof(st_table_entry*))
+#define MAX_PACKED_HASH (int)(ST_DEFAULT_PACKED_TABLE_SIZE * sizeof(st_table_entry*) / sizeof(st_packed_entry))
+
+STATIC_ASSERT(st_packed_entry, sizeof(st_packed_entry) == sizeof(st_table_entry*[PACKED_UNIT]))
+STATIC_ASSERT(st_packed_bins, sizeof(st_packed_entry[MAX_PACKED_HASH]) <= sizeof(st_table_entry*[ST_DEFAULT_PACKED_TABLE_SIZE]))
 
     /*
      * DEFAULT_MAX_DENSITY is the default for the largest we allow the
@@ -38,7 +48,8 @@ struct st_table_entry {
      *
      */
 
-static const struct st_hash_type type_numhash = {
+#define type_numhash st_hashtype_num
+const struct st_hash_type st_hashtype_num = {
     st_numcmp,
     st_numhash,
 };
@@ -61,20 +72,128 @@ static void rehash(st_table *);
 #ifdef RUBY
 #define malloc xmalloc
 #define calloc xcalloc
+#define realloc xrealloc
 #define free(x) xfree(x)
 #endif
 
 #define numberof(array) (int)(sizeof(array) / sizeof((array)[0]))
 
-#define alloc(type) (type*)malloc((size_t)sizeof(type))
-#define Calloc(n,s) (char*)calloc((n),(s))
-
 #define EQUAL(table,x,y) ((x)==(y) || (*(table)->type->compare)((x),(y)) == 0)
 
-/* remove cast to unsigned int in the future */
-#define do_hash(key,table) (unsigned int)(st_index_t)(*(table)->type->hash)((key))
+#define do_hash(key,table) (st_index_t)(*(table)->type->hash)((key))
 #define do_hash_bin(key,table) (do_hash((key), (table))%(table)->num_bins)
 
+/* preparation for possible allocation improvements */
+#ifdef POOL_ALLOC_API
+#define st_alloc_entry() (st_table_entry *)ruby_xpool_malloc_6p()
+#define st_free_entry(entry) ruby_xpool_free(entry)
+#define st_alloc_table() (st_table *)ruby_xpool_malloc_6p()
+#define st_dealloc_table(table) ruby_xpool_free(table)
+static inline st_table_entry **
+st_alloc_bins(st_index_t size)
+{
+    st_table_entry **result;
+    if (size == 11) {
+        result = (st_table_entry **) ruby_xpool_malloc_11p();
+        memset(result, 0, 11 * sizeof(st_table_entry *));
+    }
+    else
+        result = (st_table_entry **) ruby_xcalloc(size, sizeof(st_table_entry*));
+    return result;
+}
+static inline void
+st_free_bins(st_table_entry **bins, st_index_t size)
+{
+    if (size == 11)
+	ruby_xpool_free(bins);
+    else
+	ruby_xfree(bins);
+}
+static inline st_table_entry**
+st_realloc_bins(st_table_entry **bins, st_index_t newsize, st_index_t oldsize)
+{
+    st_table_entry **new_bins = st_alloc_bins(newsize);
+    st_free_bins(bins, oldsize);
+    return new_bins;
+}
+#else
+#define st_alloc_entry() (st_table_entry *)malloc(sizeof(st_table_entry))
+#define st_free_entry(entry) free(entry)
+#define st_alloc_table() (st_table *)malloc(sizeof(st_table))
+#define st_dealloc_table(table) free(table)
+#define st_alloc_bins(size) (st_table_entry **)calloc(size, sizeof(st_table_entry *))
+#define st_free_bins(bins, size) free(bins)
+static inline st_table_entry**
+st_realloc_bins(st_table_entry **bins, st_index_t newsize, st_index_t oldsize)
+{
+    bins = (st_table_entry **)realloc(bins, newsize * sizeof(st_table_entry *));
+    MEMZERO(bins, st_table_entry*, newsize);
+    return bins;
+}
+#endif
+
+/* Shortage */
+#define bins as.big.bins
+#define head as.big.head
+#define tail as.big.tail
+#define real_entries as.packed.real_entries
+
+/* preparation for possible packing improvements */
+#define PACKED_BINS(table) ((table)->as.packed.entries)
+#define PACKED_ENT(table, i) PACKED_BINS(table)[i]
+#define PKEY(table, i) PACKED_ENT((table), (i)).key
+#define PVAL(table, i) PACKED_ENT((table), (i)).val
+#define PHASH(table, i) PACKED_ENT((table), (i)).hash
+#define PKEY_SET(table, i, v) (PKEY((table), (i)) = (v))
+#define PVAL_SET(table, i, v) (PVAL((table), (i)) = (v))
+#define PHASH_SET(table, i, v) (PHASH((table), (i)) = (v))
+
+/* this function depends much on packed layout, so that it placed here */
+static inline void
+remove_packed_entry(st_table *table, st_index_t i)
+{
+    table->real_entries--;
+    table->num_entries--;
+    if (i < table->real_entries) {
+	MEMMOVE(&PACKED_ENT(table, i), &PACKED_ENT(table, i+1),
+		st_packed_entry, table->real_entries - i);
+    }
+}
+
+static inline void
+remove_safe_packed_entry(st_table *table, st_index_t i, st_data_t never)
+{
+    table->num_entries--;
+    PKEY_SET(table, i, never);
+    PVAL_SET(table, i, never);
+    PHASH_SET(table, i, 0);
+}
+
+/* ultrapacking */
+#define real_upacked num_bins
+#define MAX_UPACKED_HASH 1
+#define ULTRAPACKED(table) ((table)->real_upacked <= 1)
+#define UPACKED_ENT(table) ((table)->as.upacked)
+#define UPKEY(table) UPACKED_ENT(table).key
+#define UPVAL(table) UPACKED_ENT(table).val
+#define UPHASH(table) UPACKED_ENT(table).hash
+#define UPKEY_SET(table, v) (UPACKED_ENT(table).key = (v))
+#define UPVAL_SET(table, v) (UPACKED_ENT(table).val = (v))
+#define UPHASH_SET(table, v) (UPACKED_ENT(table).hash = (v))
+static inline void
+remove_upacked_entry(st_table *table)
+{
+    table->real_upacked = table->num_entries = 0;
+}
+
+static inline void
+remove_safe_upacked_entry(st_table *table, st_data_t never)
+{
+    table->num_entries = 0;
+    UPKEY_SET(table, never);
+    UPVAL_SET(table, never);
+    UPHASH_SET(table, 0);
+}
 /*
  * MINSIZE is the minimum size of a dictionary.
  */
@@ -85,8 +204,8 @@ static void rehash(st_table *);
 Table of prime numbers 2^n+a, 2<=n<=30.
 */
 static const unsigned int primes[] = {
-	8 + 3,
-	16 + 3,
+	ST_DEFAULT_INIT_TABLE_SIZE,
+	ST_DEFAULT_SECOND_TABLE_SIZE,
 	32 + 5,
 	64 + 3,
 	128 + 3,
@@ -161,8 +280,6 @@ stat_col(void)
 }
 #endif
 
-#define MAX_PACKED_NUMHASH (ST_DEFAULT_INIT_TABLE_SIZE/2)
-
 st_table*
 st_init_table_with_size(const struct st_hash_type *type, st_index_t size)
 {
@@ -181,14 +298,19 @@ st_init_table_with_size(const struct st_hash_type *type, st_index_t size)
     }
 #endif
 
-    size = new_size(size);	/* round up to prime number */
 
-    tbl = alloc(st_table);
+    tbl = st_alloc_table();
     tbl->type = type;
     tbl->num_entries = 0;
-    tbl->entries_packed = type == &type_numhash && size/2 <= MAX_PACKED_NUMHASH;
+    tbl->entries_packed = size <= MAX_PACKED_HASH;
+    if (tbl->entries_packed) {
+	size = size <= MAX_UPACKED_HASH ? 0 : ST_DEFAULT_PACKED_TABLE_SIZE;
+    }
+    else {
+	size = new_size(size);	/* round up to prime number */
+    }
     tbl->num_bins = size;
-    tbl->bins = (st_table_entry **)Calloc(size, sizeof(st_table_entry*));
+    tbl->bins = size ? st_alloc_bins(size) : 0;
     tbl->head = 0;
     tbl->tail = 0;
 
@@ -243,17 +365,23 @@ st_clear(st_table *table)
     register st_table_entry *ptr, *next;
     st_index_t i;
 
+    if (ULTRAPACKED(table)) {
+	remove_upacked_entry(table);
+	return;
+    }
+
     if (table->entries_packed) {
         table->num_entries = 0;
+        table->real_entries = 0;
         return;
     }
 
-    for(i = 0; i < table->num_bins; i++) {
+    for (i = 0; i < table->num_bins; i++) {
 	ptr = table->bins[i];
 	table->bins[i] = 0;
 	while (ptr != 0) {
 	    next = ptr->next;
-	    free(ptr);
+	    st_free_entry(ptr);
 	    ptr = next;
 	}
     }
@@ -266,14 +394,19 @@ void
 st_free_table(st_table *table)
 {
     st_clear(table);
-    free(table->bins);
-    free(table);
+    if (!ULTRAPACKED(table)) {
+	st_free_bins(table->bins, table->num_bins);
+    }
+    st_dealloc_table(table);
 }
 
 size_t
 st_memsize(const st_table *table)
 {
-    if (table->entries_packed) {
+    if (ULTRAPACKED(table)) {
+	return sizeof(st_table);
+    }
+    else if (table->entries_packed) {
 	return table->num_bins * sizeof (void *) + sizeof(st_table);
     }
     else {
@@ -306,46 +439,77 @@ count_collision(const struct st_hash_type *type)
 #define FOUND_ENTRY
 #endif
 
-#define FIND_ENTRY(table, ptr, hash_val, bin_pos) do {\
-    (bin_pos) = (hash_val)%(table)->num_bins;\
-    (ptr) = (table)->bins[(bin_pos)];\
-    FOUND_ENTRY;\
-    if (PTR_NOT_EQUAL((table), (ptr), (hash_val), key)) {\
-	COLLISION;\
-	while (PTR_NOT_EQUAL((table), (ptr)->next, (hash_val), key)) {\
-	    (ptr) = (ptr)->next;\
-	}\
-	(ptr) = (ptr)->next;\
-    }\
-} while (0)
+#define FIND_ENTRY(table, ptr, hash_val, bin_pos) \
+    ((ptr) = find_entry((table), key, (hash_val), ((bin_pos) = (hash_val)%(table)->num_bins)))
+
+static st_table_entry *
+find_entry(st_table *table, st_data_t key, st_index_t hash_val, st_index_t bin_pos)
+{
+    register st_table_entry *ptr = table->bins[bin_pos];
+    FOUND_ENTRY;
+    if (PTR_NOT_EQUAL(table, ptr, hash_val, key)) {
+	COLLISION;
+	while (PTR_NOT_EQUAL(table, ptr->next, hash_val, key)) {
+	    ptr = ptr->next;
+	}
+	ptr = ptr->next;
+    }
+    return ptr;
+}
+
+static inline st_index_t
+find_packed_index(st_table *table, st_index_t hash_val, st_data_t key)
+{
+    st_index_t i = 0;
+    while (i < table->real_entries &&
+	   (PHASH(table, i) != hash_val || !EQUAL(table, key, PKEY(table, i)))) {
+	i++;
+    }
+    return i;
+}
+
+static inline int
+check_upacked(st_table *table, st_index_t hash_val, st_data_t key)
+{
+    return table->num_entries &&
+	UPHASH(table) == hash_val &&
+	EQUAL(table, key, UPKEY(table));
+}
 
 #define collision_check 0
 
 int
 st_lookup(st_table *table, register st_data_t key, st_data_t *value)
 {
-    st_index_t hash_val, bin_pos;
+    st_index_t hash_val;
     register st_table_entry *ptr;
 
+    hash_val = do_hash(key, table);
+
+    if (ULTRAPACKED(table)) {
+	if (check_upacked(table, hash_val, key)) {
+	    if (value != 0) *value = UPVAL(table);
+	    return 1;
+	}
+	return 0;
+    }
+
     if (table->entries_packed) {
-        st_index_t i;
-        for (i = 0; i < table->num_entries; i++) {
-            if ((st_data_t)table->bins[i*2] == key) {
-                if (value !=0) *value = (st_data_t)table->bins[i*2+1];
-                return 1;
-            }
-        }
+	st_index_t i = find_packed_index(table, hash_val, key);
+	if (i < table->real_entries) {
+	    if (value != 0) *value = PVAL(table, i);
+	    return 1;
+	}
         return 0;
     }
 
-    hash_val = do_hash(key, table);
-    FIND_ENTRY(table, ptr, hash_val, bin_pos);
+    ptr = find_entry(table, key, hash_val, hash_val % table->num_bins);
 
     if (ptr == 0) {
 	return 0;
     }
     else {
-	if (value != 0)  *value = ptr->record;
+	if (value != 0) *value = ptr->record;
 	return 1;
     }
 }
@@ -353,22 +517,29 @@ st_lookup(st_table *table, register st_data_t key, st_data_t *value)
 int
 st_get_key(st_table *table, register st_data_t key, st_data_t *result)
 {
-    st_index_t hash_val, bin_pos;
+    st_index_t hash_val;
     register st_table_entry *ptr;
 
+    hash_val = do_hash(key, table);
+
+    if (ULTRAPACKED(table)) {
+	if (check_upacked(table, hash_val, key)) {
+	    if (result != 0) *result = UPKEY(table);
+	    return 1;
+	}
+	return 0;
+    }
+
     if (table->entries_packed) {
-        st_index_t i;
-        for (i = 0; i < table->num_entries; i++) {
-            if ((st_data_t)table->bins[i*2] == key) {
-                if (result !=0) *result = (st_data_t)table->bins[i*2];
-                return 1;
-            }
-        }
+	st_index_t i = find_packed_index(table, hash_val, key);
+	if (i < table->real_entries) {
+	    if (result != 0) *result = PKEY(table, i);
+	    return 1;
+	}
         return 0;
     }
 
-    hash_val = do_hash(key, table);
-    FIND_ENTRY(table, ptr, hash_val, bin_pos);
+    ptr = find_entry(table, key, hash_val, hash_val % table->num_bins);
 
     if (ptr == 0) {
 	return 0;
@@ -382,85 +553,151 @@ st_get_key(st_table *table, register st_data_t key, st_data_t *result)
 #undef collision_check
 #define collision_check 1
 
-#define MORE_PACKABLE_P(table) \
-    ((st_index_t)((table)->num_entries+1) * 2 <= (table)->num_bins && \
-     (table)->num_entries+1 <= MAX_PACKED_NUMHASH)
-
-#define ADD_DIRECT(table, key, value, hash_val, bin_pos)\
-do {\
-    st_table_entry *entry;\
-    if ((table)->num_entries > ST_DEFAULT_MAX_DENSITY * (table)->num_bins) {\
-	rehash(table);\
-        (bin_pos) = (hash_val) % (table)->num_bins;\
-    }\
-    \
-    entry = alloc(st_table_entry);\
-    \
-    entry->hash = (hash_val);\
-    entry->key = (key);\
-    entry->record = (value);\
-    entry->next = (table)->bins[(bin_pos)];\
-    if ((table)->head != 0) {\
-	entry->fore = 0;\
-	(entry->back = (table)->tail)->fore = entry;\
-	(table)->tail = entry;\
-    }\
-    else {\
-	(table)->head = (table)->tail = entry;\
-	entry->fore = entry->back = 0;\
-    }\
-    (table)->bins[(bin_pos)] = entry;\
-    (table)->num_entries++;\
-} while (0)
+static inline st_table_entry *
+new_entry(st_table * table, st_data_t key, st_data_t value,
+	st_index_t hash_val, register st_index_t bin_pos)
+{
+    register st_table_entry *entry = st_alloc_entry();
+
+    entry->next = table->bins[bin_pos];
+    table->bins[bin_pos] = entry;
+    entry->hash = hash_val;
+    entry->key = key;
+    entry->record = value;
+
+    return entry;
+}
+
+static inline void
+add_direct(st_table *table, st_data_t key, st_data_t value,
+	   st_index_t hash_val, register st_index_t bin_pos)
+{
+    register st_table_entry *entry;
+    if (table->num_entries > ST_DEFAULT_MAX_DENSITY * table->num_bins) {
+	rehash(table);
+        bin_pos = hash_val % table->num_bins;
+    }
+
+    entry = new_entry(table, key, value, hash_val, bin_pos);
+
+    if (table->head != 0) {
+	entry->fore = 0;
+	(entry->back = table->tail)->fore = entry;
+	table->tail = entry;
+    }
+    else {
+	table->head = table->tail = entry;
+	entry->fore = entry->back = 0;
+    }
+    table->num_entries++;
+}
 
 static void
 unpack_entries(register st_table *table)
 {
     st_index_t i;
-    struct st_table_entry *packed_bins[MAX_PACKED_NUMHASH*2];
+    st_packed_entry packed_bins[MAX_PACKED_HASH];
+    register st_table_entry *entry, *preventry = 0, **chain;
     st_table tmp_table = *table;
 
-    memcpy(packed_bins, table->bins, sizeof(struct st_table_entry *) * table->num_entries*2);
-    table->bins = packed_bins;
+    MEMCPY(packed_bins, PACKED_BINS(table), st_packed_entry, MAX_PACKED_HASH);
+    table->as.packed.entries = packed_bins;
     tmp_table.entries_packed = 0;
-    tmp_table.num_entries = 0;
-    memset(tmp_table.bins, 0, sizeof(struct st_table_entry *) * tmp_table.num_bins);
-    for (i = 0; i < table->num_entries; i++) {
-        st_insert(&tmp_table, (st_data_t)packed_bins[i*2], (st_data_t)packed_bins[i*2+1]);
-    }
+#if ST_DEFAULT_INIT_TABLE_SIZE == ST_DEFAULT_PACKED_TABLE_SIZE
+    MEMZERO(tmp_table.bins, st_table_entry*, tmp_table.num_bins);
+#else
+    tmp_table.bins = st_realloc_bins(tmp_table.bins, ST_DEFAULT_INIT_TABLE_SIZE, tmp_table.num_bins);
+    tmp_table.num_bins = ST_DEFAULT_INIT_TABLE_SIZE;
+#endif
+    i = 0;
+    chain = &tmp_table.head;
+    do {
+	st_data_t key = packed_bins[i].key;
+	st_data_t val = packed_bins[i].val;
+	st_index_t hash = packed_bins[i].hash;
+	entry = new_entry(&tmp_table, key, val, hash,
+			  hash % ST_DEFAULT_INIT_TABLE_SIZE);
+	*chain = entry;
+	entry->back = preventry;
+	preventry = entry;
+	chain = &entry->fore;
+    } while (++i < MAX_PACKED_HASH);
+    *chain = NULL;
+    tmp_table.tail = entry;
     *table = tmp_table;
 }
 
+static void
+add_packed_direct(st_table *table, st_data_t key, st_data_t value, st_index_t hash_val)
+{
+    if (table->real_entries < MAX_PACKED_HASH) {
+	st_index_t i = table->real_entries++;
+	PKEY_SET(table, i, key);
+	PVAL_SET(table, i, value);
+	PHASH_SET(table, i, hash_val);
+	table->num_entries++;
+    }
+    else {
+	unpack_entries(table);
+	add_direct(table, key, value, hash_val, hash_val % table->num_bins);
+    }
+}
+
+static void
+add_upacked_direct(register st_table *table, register st_data_t key, st_data_t value, st_index_t hash_val)
+{
+    if (table->real_upacked) {
+	st_packed_entry *entries = (st_packed_entry *) st_alloc_bins(ST_DEFAULT_PACKED_TABLE_SIZE);
+	entries[0] = UPACKED_ENT(table);
+	entries[1].hash = hash_val;
+	entries[1].key = key;
+	entries[1].val = value;
+	table->num_bins = ST_DEFAULT_PACKED_TABLE_SIZE;
+	table->real_entries = 2;
+	table->num_entries++;
+	table->as.packed.entries = entries;
+    }
+    else {
+	table->real_upacked = 1;
+	table->num_entries = 1;
+	UPHASH_SET(table, hash_val);
+	UPKEY_SET(table, key);
+	UPVAL_SET(table, value);
+    }
+}
+
 int
 st_insert(register st_table *table, register st_data_t key, st_data_t value)
 {
-    st_index_t hash_val, bin_pos;
+    st_index_t hash_val;
+    register st_index_t bin_pos;
     register st_table_entry *ptr;
 
+    hash_val = do_hash(key, table);
+
+    if (ULTRAPACKED(table)) {
+	if (check_upacked(table, hash_val, key)) {
+	    UPVAL_SET(table, value);
+	    return 1;
+	}
+	add_upacked_direct(table, key, value, hash_val);
+	return 0;
+    }
+
     if (table->entries_packed) {
-        st_index_t i;
-        for (i = 0; i < table->num_entries; i++) {
-            if ((st_data_t)table->bins[i*2] == key) {
-                table->bins[i*2+1] = (struct st_table_entry*)value;
-                return 1;
-            }
-        }
-        if (MORE_PACKABLE_P(table)) {
-            i = table->num_entries++;
-            table->bins[i*2] = (struct st_table_entry*)key;
-            table->bins[i*2+1] = (struct st_table_entry*)value;
-            return 0;
-        }
-        else {
-            unpack_entries(table);
+	st_index_t i = find_packed_index(table, hash_val, key);
+	if (i < table->real_entries) {
+	    PVAL_SET(table, i, value);
+	    return 1;
         }
+	add_packed_direct(table, key, value, hash_val);
+	return 0;
     }
 
-    hash_val = do_hash(key, table);
     FIND_ENTRY(table, ptr, hash_val, bin_pos);
 
     if (ptr == 0) {
-	ADD_DIRECT(table, key, value, hash_val, bin_pos);
+	add_direct(table, key, value, hash_val, bin_pos);
 	return 0;
     }
     else {
@@ -473,34 +710,38 @@ int
 st_insert2(register st_table *table, register st_data_t key, st_data_t value,
 	   st_data_t (*func)(st_data_t))
 {
-    st_index_t hash_val, bin_pos;
+    st_index_t hash_val;
+    register st_index_t bin_pos;
     register st_table_entry *ptr;
 
+    hash_val = do_hash(key, table);
+
+    if (ULTRAPACKED(table)) {
+	if (check_upacked(table, hash_val, key)) {
+	    UPVAL_SET(table, value);
+	    return 1;
+	}
+	key = (*func)(key);
+	add_upacked_direct(table, key, value, hash_val);
+	return 0;
+    }
+
     if (table->entries_packed) {
-        st_index_t i;
-        for (i = 0; i < table->num_entries; i++) {
-            if ((st_data_t)table->bins[i*2] == key) {
-                table->bins[i*2+1] = (struct st_table_entry*)value;
-                return 1;
-            }
-        }
-        if (MORE_PACKABLE_P(table)) {
-            i = table->num_entries++;
-            table->bins[i*2] = (struct st_table_entry*)key;
-            table->bins[i*2+1] = (struct st_table_entry*)value;
-            return 0;
-        }
-        else {
-            unpack_entries(table);
-        }
+	st_index_t i = find_packed_index(table, hash_val, key);
+	if (i < table->real_entries) {
+	    PVAL_SET(table, i, value);
+	    return 1;
+	}
+	key = (*func)(key);
+	add_packed_direct(table, key, value, hash_val);
+	return 0;
     }
 
-    hash_val = do_hash(key, table);
     FIND_ENTRY(table, ptr, hash_val, bin_pos);
 
     if (ptr == 0) {
 	key = (*func)(key);
-	ADD_DIRECT(table, key, value, hash_val, bin_pos);
+	add_direct(table, key, value, hash_val, bin_pos);
 	return 0;
     }
     else {
@@ -512,36 +753,30 @@ st_insert2(register st_table *table, register st_data_t key, st_data_t value,
 void
 st_add_direct(st_table *table, st_data_t key, st_data_t value)
 {
-    st_index_t hash_val, bin_pos;
+    st_index_t hash_val;
+
+    hash_val = do_hash(key, table);
+    if (ULTRAPACKED(table)) {
+	add_upacked_direct(table, key, value, hash_val);
+	return;
+    }
 
     if (table->entries_packed) {
-        int i;
-        if (MORE_PACKABLE_P(table)) {
-            i = table->num_entries++;
-            table->bins[i*2] = (struct st_table_entry*)key;
-            table->bins[i*2+1] = (struct st_table_entry*)value;
-            return;
-        }
-        else {
-            unpack_entries(table);
-        }
+	add_packed_direct(table, key, value, hash_val);
+	return;
     }
 
-    hash_val = do_hash(key, table);
-    bin_pos = hash_val % table->num_bins;
-    ADD_DIRECT(table, key, value, hash_val, bin_pos);
+    add_direct(table, key, value, hash_val, hash_val % table->num_bins);
 }
 
 static void
 rehash(register st_table *table)
 {
     register st_table_entry *ptr, **new_bins;
-    st_index_t i, new_num_bins, hash_val;
+    st_index_t new_num_bins, hash_val;
 
     new_num_bins = new_size(table->num_bins+1);
-    new_bins = (st_table_entry**)
-	xrealloc(table->bins, new_num_bins * sizeof(st_table_entry*));
-    for (i = 0; i < new_num_bins; ++i) new_bins[i] = 0;
+    new_bins = st_realloc_bins(table->bins, new_num_bins, table->num_bins);
     table->num_bins = new_num_bins;
     table->bins = new_bins;
 
@@ -558,34 +793,37 @@ st_table*
 st_copy(st_table *old_table)
 {
     st_table *new_table;
-    st_table_entry *ptr, *entry, *prev, **tail;
+    st_table_entry *ptr, *entry, *prev, **tailp;
     st_index_t num_bins = old_table->num_bins;
     st_index_t hash_val;
 
-    new_table = alloc(st_table);
+    new_table = st_alloc_table();
     if (new_table == 0) {
 	return 0;
     }
 
     *new_table = *old_table;
-    new_table->bins = (st_table_entry**)
-	Calloc((unsigned)num_bins, sizeof(st_table_entry*));
+    if (ULTRAPACKED(old_table)) {
+	return new_table;
+    }
+
+    new_table->bins = st_alloc_bins(num_bins);
 
     if (new_table->bins == 0) {
-	free(new_table);
+	st_dealloc_table(new_table);
 	return 0;
     }
 
     if (old_table->entries_packed) {
-        memcpy(new_table->bins, old_table->bins, sizeof(struct st_table_entry *) * old_table->num_bins);
+        MEMCPY(new_table->bins, old_table->bins, st_table_entry*, old_table->num_bins);
         return new_table;
     }
 
     if ((ptr = old_table->head) != 0) {
 	prev = 0;
-	tail = &new_table->head;
+	tailp = &new_table->head;
 	do {
-	    entry = alloc(st_table_entry);
+	    entry = st_alloc_entry();
 	    if (entry == 0) {
 		st_free_table(new_table);
 		return 0;
@@ -595,8 +833,8 @@ st_copy(st_table *old_table)
 	    entry->next = new_table->bins[hash_val];
 	    new_table->bins[hash_val] = entry;
 	    entry->back = prev;
-	    *tail = prev = entry;
-	    tail = &entry->fore;
+	    *tailp = prev = entry;
+	    tailp = &entry->fore;
 	} while ((ptr = ptr->fore) != 0);
 	new_table->tail = prev;
     }
@@ -604,21 +842,22 @@ st_copy(st_table *old_table)
     return new_table;
 }
 
-#define REMOVE_ENTRY(table, ptr) do					\
-    {									\
-	if ((ptr)->fore == 0 && (ptr)->back == 0) {			\
-	    (table)->head = 0;						\
-	    (table)->tail = 0;						\
-	}								\
-	else {								\
-	    st_table_entry *fore = (ptr)->fore, *back = (ptr)->back;	\
-	    if (fore) fore->back = back;				\
-	    if (back) back->fore = fore;				\
-	    if ((ptr) == (table)->head) (table)->head = fore;		\
-	    if ((ptr) == (table)->tail) (table)->tail = back;		\
-	}								\
-	(table)->num_entries--;						\
-    } while (0)
+static inline void
+remove_entry(st_table *table, st_table_entry *ptr)
+{
+    if (ptr->fore == 0 && ptr->back == 0) {
+	table->head = 0;
+	table->tail = 0;
+    }
+    else {
+	st_table_entry *fore = ptr->fore, *back = ptr->back;
+	if (fore) fore->back = back;
+	if (back) back->fore = fore;
+	if (ptr == table->head) table->head = fore;
+	if (ptr == table->tail) table->tail = back;
+    }
+    table->num_entries--;
+}
 
 int
 st_delete(register st_table *table, register st_data_t *key, st_data_t *value)
@@ -627,30 +866,38 @@ st_delete(register st_table *table, register st_data_t *key, st_data_t *value)
     st_table_entry **prev;
     register st_table_entry *ptr;
 
+    hash_val = do_hash(*key, table);
+
+    if (ULTRAPACKED(table)) {
+	if (check_upacked(table, hash_val, *key)) {
+	    if (value != 0) *value = UPVAL(table);
+	    *key = UPKEY(table);
+	    remove_upacked_entry(table);
+	    return 1;
+	}
+	return 0;
+    }
+
     if (table->entries_packed) {
-        st_index_t i;
-        for (i = 0; i < table->num_entries; i++) {
-            if ((st_data_t)table->bins[i*2] == *key) {
-                if (value != 0) *value = (st_data_t)table->bins[i*2+1];
-                table->num_entries--;
-                memmove(&table->bins[i*2], &table->bins[(i+1)*2],
-                        sizeof(struct st_table_entry*) * 2*(table->num_entries-i));
-                return 1;
-            }
+	st_index_t i = find_packed_index(table, hash_val, *key);
+	if (i < table->real_entries) {
+	    if (value != 0) *value = PVAL(table, i);
+	    *key = PKEY(table, i);
+	    remove_packed_entry(table, i);
+	    return 1;
         }
         if (value != 0) *value = 0;
         return 0;
     }
 
-    hash_val = do_hash_bin(*key, table);
-
-    for (prev = &table->bins[hash_val]; (ptr = *prev) != 0; prev = &ptr->next) {
+    prev = &table->bins[hash_val % table->num_bins];
+    for (;(ptr = *prev) != 0; prev = &ptr->next) {
 	if (EQUAL(table, *key, ptr->key)) {
 	    *prev = ptr->next;
-	    REMOVE_ENTRY(table, ptr);
+	    remove_entry(table, ptr);
 	    if (value != 0) *value = ptr->record;
 	    *key = ptr->key;
-	    free(ptr);
+	    st_free_entry(ptr);
 	    return 1;
 	}
     }
@@ -665,25 +912,36 @@ st_delete_safe(register st_table *table, register st_data_t *key, st_data_t *val
     st_index_t hash_val;
     register st_table_entry *ptr;
 
+    hash_val = do_hash(*key, table);
+
+    if (ULTRAPACKED(table)) {
+	if (check_upacked(table, hash_val, *key)) {
+	    if (value != 0) *value = UPVAL(table);
+	    *key = UPKEY(table);
+	    remove_safe_upacked_entry(table, never);
+	    return 1;
+	}
+	if (value != 0) *value = 0;
+	return 0;
+    }
+
     if (table->entries_packed) {
-	st_index_t i;
-	for (i = 0; i < table->num_entries; i++) {
-	    if ((st_data_t)table->bins[i*2] == *key) {
-		if (value != 0) *value = (st_data_t)table->bins[i*2+1];
-		table->bins[i*2] = (void *)never;
-		return 1;
-	    }
+	st_index_t i = find_packed_index(table, hash_val, *key);
+	if (i < table->real_entries) {
+	    if (value != 0) *value = PVAL(table, i);
+	    *key = PKEY(table, i);
+	    remove_safe_packed_entry(table, i, never);
+	    return 1;
 	}
 	if (value != 0) *value = 0;
 	return 0;
     }
 
-    hash_val = do_hash_bin(*key, table);
-    ptr = table->bins[hash_val];
+    ptr = table->bins[hash_val % table->num_bins];
 
     for (; ptr != 0; ptr = ptr->next) {
 	if ((ptr->key != never) && EQUAL(table, ptr->key, *key)) {
-	    REMOVE_ENTRY(table, ptr);
+	    remove_entry(table, ptr);
 	    *key = ptr->key;
 	    if (value != 0) *value = ptr->record;
 	    ptr->key = ptr->record = never;
@@ -701,17 +959,23 @@ st_cleanup_safe(st_table *table, st_data_t never)
     st_table_entry *ptr, **last, *tmp;
     st_index_t i;
 
+    if (ULTRAPACKED(table)) {
+	table->real_upacked = table->num_entries;
+	return;
+    }
+
     if (table->entries_packed) {
 	st_index_t i = 0, j = 0;
-	while ((st_data_t)table->bins[i*2] != never) {
-	    if (i++ == table->num_entries) return;
+	while (PKEY(table, i) != never) {
+	    if (i++ == table->real_entries) return;
 	}
-	for (j = i; ++i < table->num_entries;) {
-	    if ((st_data_t)table->bins[i*2] == never) continue;
-	    table->bins[j*2] = table->bins[i*2];
-	    table->bins[j*2+1] = table->bins[i*2+1];
+	for (j = i; ++i < table->real_entries;) {
+	    if (PKEY(table, i) == never) continue;
+	    PACKED_ENT(table, j) = PACKED_ENT(table, i);
 	    j++;
 	}
+	table->real_entries = j;
+	/* table->num_entries really should be equal j at this moment, but let set it anyway */
 	table->num_entries = j;
 	return;
     }
@@ -722,7 +986,7 @@ st_cleanup_safe(st_table *table, st_data_t never)
 	    if (ptr->key == never) {
 		tmp = ptr;
 		*last = ptr = ptr->next;
-		free(tmp);
+		st_free_entry(tmp);
 	    }
 	    else {
 		ptr = *(last = &ptr->next);
@@ -732,50 +996,78 @@ st_cleanup_safe(st_table *table, st_data_t never)
 }
 
 int
-st_foreach(st_table *table, int (*func)(ANYARGS), st_data_t arg)
+st_foreach_check(st_table *table, int (*func)(ANYARGS), st_data_t arg, st_data_t never)
 {
     st_table_entry *ptr, **last, *tmp;
     enum st_retval retval;
-    st_index_t i;
+    st_data_t key, val;
+    st_index_t hash, i = 0;
+
+    if (table->num_entries == 0) {
+	return 0;
+    }
+
+    if (ULTRAPACKED(table)) {
+	key = UPKEY(table);
+	val = UPVAL(table);
+	hash = UPHASH(table);
+	if (key == never) return 0;
+	retval = (*func)(key, val, arg);
+	if (!ULTRAPACKED(table)) {
+	    goto packed;
+	}
+	switch(retval) {
+	  case ST_CHECK:
+	    if (UPHASH(table) == 0 && UPKEY(table) == never)
+		break;
+	    if (check_upacked(table, hash, key))
+		break;
+	    goto deleted;
+	  case ST_DELETE:
+	    remove_safe_upacked_entry(table, never);
+	  case ST_CONTINUE:
+	  case ST_STOP:
+	    break;
+	}
+	return 0;
+    }
 
     if (table->entries_packed) {
-        for (i = 0; i < table->num_entries; i++) {
-            st_index_t j;
-            st_data_t key, val;
-            key = (st_data_t)table->bins[i*2];
-            val = (st_data_t)table->bins[i*2+1];
-            retval = (*func)(key, val, arg);
+	for (i = 0; i < table->real_entries; i++) {
+	    key = PKEY(table, i);
+	    val = PVAL(table, i);
+	    hash = PHASH(table, i);
+	    if (key == never) continue;
+	    retval = (*func)(key, val, arg);
+	  packed:
 	    if (!table->entries_packed) {
-		FIND_ENTRY(table, ptr, key, i);
+		FIND_ENTRY(table, ptr, hash, i);
 		if (retval == ST_CHECK) {
 		    if (!ptr) goto deleted;
 		    goto unpacked_continue;
 		}
 		goto unpacked;
 	    }
-            switch (retval) {
+	    switch (retval) {
 	      case ST_CHECK:	/* check if hash is modified during iteration */
-                for (j = 0; j < table->num_entries; j++) {
-                    if ((st_data_t)table->bins[j*2] == key)
-                        break;
-                }
-                if (j == table->num_entries) {
+		if (PHASH(table, i) == 0 && PKEY(table, i) == never) {
+		    break;
+		}
+		i = find_packed_index(table, hash, key);
+		if (i == table->real_entries) {
 		    goto deleted;
-                }
+		}
 		/* fall through */
 	      case ST_CONTINUE:
 		break;
 	      case ST_STOP:
 		return 0;
 	      case ST_DELETE:
-                table->num_entries--;
-                memmove(&table->bins[i*2], &table->bins[(i+1)*2],
-                        sizeof(struct st_table_entry*) * 2*(table->num_entries-i));
-                i--;
-                break;
-            }
-        }
-        return 0;
+		remove_safe_packed_entry(table, i, never);
+		break;
+	    }
+	}
+	return 0;
     }
     else {
 	ptr = table->head;
@@ -783,6 +1075,8 @@ st_foreach(st_table *table, int (*func)(ANYARGS), st_data_t arg)
 
     if (ptr != 0) {
 	do {
+	    if (ptr->key == never)
+		goto unpacked_continue;
 	    i = ptr->hash % table->num_bins;
 	    retval = (*func)(ptr->key, ptr->record, arg);
 	  unpacked:
@@ -808,10 +1102,100 @@ st_foreach(st_table *table, int (*func)(ANYARGS), st_data_t arg)
 		for (; (tmp = *last) != 0; last = &tmp->next) {
 		    if (ptr == tmp) {
 			tmp = ptr->fore;
+			remove_entry(table, ptr);
+			ptr->key = ptr->record = never;
+			ptr->hash = 0;
+			ptr = tmp;
+			break;
+		    }
+		}
+	    }
+	} while (ptr && table->head);
+    }
+    return 0;
+}
+
+int
+st_foreach(st_table *table, int (*func)(ANYARGS), st_data_t arg)
+{
+    st_table_entry *ptr, **last, *tmp;
+    enum st_retval retval;
+    st_data_t key, val;
+    st_index_t hash, i = 0;
+
+    if (table->num_entries == 0) {
+	return 0;
+    }
+
+    if (ULTRAPACKED(table)) {
+	key = UPKEY(table);
+	val = UPVAL(table);
+	hash = UPHASH(table);
+	retval = (*func)(key, val, arg);
+	if (!ULTRAPACKED(table)) {
+	    goto packed;
+	}
+	switch (retval) {
+	  case ST_DELETE:
+	    remove_upacked_entry(table);
+	  case ST_CONTINUE:
+	  case ST_CHECK:
+	  case ST_STOP:
+	    break;
+	}
+	return 0;
+    }
+
+    if (table->entries_packed) {
+	for (i = 0; i < table->real_entries; i++) {
+	    key = PKEY(table, i);
+	    val = PVAL(table, i);
+	    hash = PHASH(table, i);
+	    retval = (*func)(key, val, arg);
+	  packed:
+	    if (!table->entries_packed) {
+		FIND_ENTRY(table, ptr, hash, i);
+		if (!ptr) return 0;
+		goto unpacked;
+	    }
+	    switch (retval) {
+	      case ST_CONTINUE:
+		break;
+	      case ST_CHECK:
+	      case ST_STOP:
+		return 0;
+	      case ST_DELETE:
+		remove_packed_entry(table, i);
+		i--;
+		break;
+	    }
+	}
+	return 0;
+    }
+    else {
+	ptr = table->head;
+    }
+
+    if (ptr != 0) {
+	do {
+	    i = ptr->hash % table->num_bins;
+	    retval = (*func)(ptr->key, ptr->record, arg);
+	  unpacked:
+	    switch (retval) {
+	      case ST_CONTINUE:
+		ptr = ptr->fore;
+		break;
+	      case ST_CHECK:
+	      case ST_STOP:
+		return 0;
+	      case ST_DELETE:
+		last = &table->bins[ptr->hash % table->num_bins];
+		for (; (tmp = *last) != 0; last = &tmp->next) {
+		    if (ptr == tmp) {
+			tmp = ptr->fore;
 			*last = ptr->next;
-			REMOVE_ENTRY(table, ptr);
-			free(ptr);
-			if (ptr == tmp) return 0;
+			remove_entry(table, ptr);
+			st_free_entry(ptr);
 			ptr = tmp;
 			break;
 		    }
@@ -834,13 +1218,13 @@ st_reverse_foreach(st_table *table, int (*func)(ANYARGS), st_data_t arg)
         for (i = table->num_entries-1; 0 <= i; i--) {
             int j;
             st_data_t key, val;
-            key = (st_data_t)table->bins[i*2];
-            val = (st_data_t)table->bins[i*2+1];
+            key = PKEY(table, i);
+            val = PVAL(table, i);
             retval = (*func)(key, val, arg);
             switch (retval) {
 	      case ST_CHECK:	/* check if hash is modified during iteration */
                 for (j = 0; j < table->num_entries; j++) {
-                    if ((st_data_t)table->bins[j*2] == key)
+                    if (PKEY(table, j) == key)
                         break;
                 }
                 if (j == table->num_entries) {
@@ -854,9 +1238,7 @@ st_reverse_foreach(st_table *table, int (*func)(ANYARGS), st_data_t arg)
 	      case ST_STOP:
 		return 0;
 	      case ST_DELETE:
-                table->num_entries--;
-                memmove(&table->bins[i*2], &table->bins[(i+1)*2],
-                        sizeof(struct st_table_entry*) * 2*(table->num_entries-i));
+		remove_packed_entry(table, i);
                 break;
             }
         }
@@ -889,8 +1271,8 @@ st_reverse_foreach(st_table *table, int (*func)(ANYARGS), st_data_t arg)
 		    if (ptr == tmp) {
 			tmp = ptr->back;
 			*last = ptr->next;
-			REMOVE_ENTRY(table, ptr);
-			free(ptr);
+			remove_entry(table, ptr);
+			st_free_entry(ptr);
 			ptr = tmp;
 			break;
 		    }

03-sparse_array.diff

diff --git a/class.c b/class.c
index 450c334..6e97cb0 100644
--- a/class.c
+++ b/class.c