[pull] master from ruby:master

common.mk

@@ -1295,7 +1295,7 @@ preludes: {$(VPATH)}miniprelude.c
 
 {$(srcdir)}.rb.rbinc:
 	$(ECHO) making $@
-	-$(Q) $(MAKE) $(DUMP_AST)
+	-$(Q) $(MAKE) $(DUMP_AST_TARGET)
 	$(Q) $(BASERUBY) $(tooldir)/mk_builtin_loader.rb $(DUMP_AST) $(SRC_FILE)
 
 $(BUILTIN_BINARY:yes=built)in_binary.rbbin: $(PREP) $(BUILTIN_RB_SRCS) $(srcdir)/template/builtin_binary.rbbin.tmpl

configure.ac

@@ -113,9 +113,10 @@ AC_SUBST(HAVE_BASERUBY)
 
 AC_ARG_WITH(dump-ast,
 	AS_HELP_STRING([--with-dump-ast=DUMP_AST], [use DUMP_AST as dump_ast; for cross-compiling with a host-built dump_ast]),
-	[DUMP_AST=$withval],
-	[DUMP_AST='./dump_ast$(EXEEXT)'])
-AC_SUBST(DUMP_AST)
+	[DUMP_AST=$withval DUMP_AST_TARGET=no],
+	[DUMP_AST='./dump_ast$(EXEEXT)' DUMP_AST_TARGET='$(DUMP_AST)'])
+AC_SUBST(X_DUMP_AST, "${DUMP_AST}")
+AC_SUBST(X_DUMP_AST_TARGET, "${DUMP_AST_TARGET}")
 
 : ${GIT=git}
 HAVE_GIT=yes

depend

@@ -11654,6 +11654,7 @@ prism/node.$(OBJEXT): {$(VPATH)}prism_xallocator.h
 prism/options.$(OBJEXT): $(top_srcdir)/prism/defines.h
 prism/options.$(OBJEXT): $(top_srcdir)/prism/options.c
 prism/options.$(OBJEXT): $(top_srcdir)/prism/options.h
+prism/options.$(OBJEXT): $(top_srcdir)/prism/util/pm_arena.h
 prism/options.$(OBJEXT): $(top_srcdir)/prism/util/pm_char.h
 prism/options.$(OBJEXT): $(top_srcdir)/prism/util/pm_line_offset_list.h
 prism/options.$(OBJEXT): $(top_srcdir)/prism/util/pm_string.h
@@ -11965,6 +11966,7 @@ prism/util/pm_arena.$(OBJEXT): {$(VPATH)}internal/has/warning.h
 prism/util/pm_arena.$(OBJEXT): {$(VPATH)}internal/xmalloc.h
 prism/util/pm_arena.$(OBJEXT): {$(VPATH)}prism_xallocator.h
 prism/util/pm_buffer.$(OBJEXT): $(top_srcdir)/prism/defines.h
+prism/util/pm_buffer.$(OBJEXT): $(top_srcdir)/prism/util/pm_arena.h
 prism/util/pm_buffer.$(OBJEXT): $(top_srcdir)/prism/util/pm_buffer.c
 prism/util/pm_buffer.$(OBJEXT): $(top_srcdir)/prism/util/pm_buffer.h
 prism/util/pm_buffer.$(OBJEXT): $(top_srcdir)/prism/util/pm_char.h
@@ -11994,6 +11996,7 @@ prism/util/pm_buffer.$(OBJEXT): {$(VPATH)}internal/has/warning.h
 prism/util/pm_buffer.$(OBJEXT): {$(VPATH)}internal/xmalloc.h
 prism/util/pm_buffer.$(OBJEXT): {$(VPATH)}prism_xallocator.h
 prism/util/pm_char.$(OBJEXT): $(top_srcdir)/prism/defines.h
+prism/util/pm_char.$(OBJEXT): $(top_srcdir)/prism/util/pm_arena.h
 prism/util/pm_char.$(OBJEXT): $(top_srcdir)/prism/util/pm_char.c
 prism/util/pm_char.$(OBJEXT): $(top_srcdir)/prism/util/pm_char.h
 prism/util/pm_char.$(OBJEXT): $(top_srcdir)/prism/util/pm_line_offset_list.h
@@ -12050,6 +12053,7 @@ prism/util/pm_constant_pool.$(OBJEXT): {$(VPATH)}internal/has/warning.h
 prism/util/pm_constant_pool.$(OBJEXT): {$(VPATH)}internal/xmalloc.h
 prism/util/pm_constant_pool.$(OBJEXT): {$(VPATH)}prism_xallocator.h
 prism/util/pm_integer.$(OBJEXT): $(top_srcdir)/prism/defines.h
+prism/util/pm_integer.$(OBJEXT): $(top_srcdir)/prism/util/pm_arena.h
 prism/util/pm_integer.$(OBJEXT): $(top_srcdir)/prism/util/pm_buffer.h
 prism/util/pm_integer.$(OBJEXT): $(top_srcdir)/prism/util/pm_char.h
 prism/util/pm_integer.$(OBJEXT): $(top_srcdir)/prism/util/pm_integer.c
@@ -12080,6 +12084,7 @@ prism/util/pm_integer.$(OBJEXT): {$(VPATH)}internal/has/warning.h
 prism/util/pm_integer.$(OBJEXT): {$(VPATH)}internal/xmalloc.h
 prism/util/pm_integer.$(OBJEXT): {$(VPATH)}prism_xallocator.h
 prism/util/pm_line_offset_list.$(OBJEXT): $(top_srcdir)/prism/defines.h
+prism/util/pm_line_offset_list.$(OBJEXT): $(top_srcdir)/prism/util/pm_arena.h
 prism/util/pm_line_offset_list.$(OBJEXT): $(top_srcdir)/prism/util/pm_line_offset_list.c
 prism/util/pm_line_offset_list.$(OBJEXT): $(top_srcdir)/prism/util/pm_line_offset_list.h
 prism/util/pm_line_offset_list.$(OBJEXT): {$(VPATH)}config.h

enumerator.c

@@ -2772,7 +2772,7 @@ lazy_with_index(int argc, VALUE *argv, VALUE obj)
 }
 
 static struct MEMO *
-lazy_tee_proc(VALUE proc_entry, struct MEMO *result, VALUE memos, long memo_index)
+lazy_tap_each_proc(VALUE proc_entry, struct MEMO *result, VALUE memos, long memo_index)
 {
     struct proc_entry *entry = proc_entry_ptr(proc_entry);
 
@@ -2781,13 +2781,13 @@ lazy_tee_proc(VALUE proc_entry, struct MEMO *result, VALUE memos, long memo_inde
     return result;
 }
 
-static const lazyenum_funcs lazy_tee_funcs = {
-    lazy_tee_proc, 0,
+static const lazyenum_funcs lazy_tap_each_funcs = {
+    lazy_tap_each_proc, 0,
 };
 
 /*
  *  call-seq:
- *     lazy.tee { |item| ... } -> lazy_enumerator
+ *     lazy.tap_each { |item| ... } -> lazy_enumerator
  *
  *  Passes each element through to the block for side effects only,
  *  without modifying the element or affecting the enumeration.
@@ -2797,7 +2797,7 @@ static const lazyenum_funcs lazy_tee_funcs = {
  *  without breaking laziness or misusing +map+.
  *
  *     (1..).lazy
- *           .tee { |x| puts "got #{x}" }
+ *           .tap_each { |x| puts "got #{x}" }
  *           .select(&:even?)
  *           .first(3)
  *     # prints: got 1, got 2, ..., got 6
@@ -2807,14 +2807,14 @@ static const lazyenum_funcs lazy_tee_funcs = {
  */
 
 static VALUE
-lazy_tee(VALUE obj)
+lazy_tap_each(VALUE obj)
 {
     if (!rb_block_given_p())
     {
-        rb_raise(rb_eArgError, "tried to call lazy tee without a block");
+        rb_raise(rb_eArgError, "tried to call lazy tap_each without a block");
     }
 
-    return lazy_add_method(obj, 0, 0, Qnil, Qnil, &lazy_tee_funcs);
+    return lazy_add_method(obj, 0, 0, Qnil, Qnil, &lazy_tap_each_funcs);
 }
 
 #if 0 /* for RDoc */
@@ -4692,7 +4692,7 @@ InitVM_Enumerator(void)
     rb_define_method(rb_cLazy, "uniq", lazy_uniq, 0);
     rb_define_method(rb_cLazy, "compact", lazy_compact, 0);
     rb_define_method(rb_cLazy, "with_index", lazy_with_index, -1);
-    rb_define_method(rb_cLazy, "tee", lazy_tee, 0);
+    rb_define_method(rb_cLazy, "tap_each", lazy_tap_each, 0);
 
     lazy_use_super_method = rb_hash_new_with_size(18);
     rb_hash_aset(lazy_use_super_method, sym("map"), sym("_enumerable_map"));

prism/defines.h

@@ -91,6 +91,18 @@
 #   define inline __inline
 #endif
 
+/**
+ * Force a function to be inlined at every call site. Use sparingly — only for
+ * small, hot functions where the compiler's heuristics fail to inline.
+ */
+#if defined(_MSC_VER)
+#   define PRISM_FORCE_INLINE __forceinline
+#elif defined(__GNUC__) || defined(__clang__)
+#   define PRISM_FORCE_INLINE inline __attribute__((always_inline))
+#else
+#   define PRISM_FORCE_INLINE inline
+#endif
+
 /**
  * Old Visual Studio versions before 2015 do not implement sprintf, but instead
  * implement _snprintf. We standard that here.
@@ -264,6 +276,49 @@
     #define PRISM_UNLIKELY(x) (x)
 #endif
 
+/**
+ * Platform detection for SIMD / fast-path implementations. At most one of
+ * these macros is defined, selecting the best available vectorization strategy.
+ */
+#if (defined(__aarch64__) && defined(__ARM_NEON)) || (defined(_MSC_VER) && defined(_M_ARM64))
+    #define PRISM_HAS_NEON
+#elif (defined(__x86_64__) && defined(__SSSE3__)) || (defined(_MSC_VER) && defined(_M_X64))
+    #define PRISM_HAS_SSSE3
+#elif defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    #define PRISM_HAS_SWAR
+#endif
+
+/**
+ * Count trailing zero bits in a 64-bit value. Used by SWAR identifier scanning
+ * to find the first non-matching byte in a word.
+ *
+ * Precondition: v must be nonzero. The result is undefined when v == 0
+ * (matching the behavior of __builtin_ctzll and _BitScanForward64).
+ */
+#if defined(__GNUC__) || defined(__clang__)
+    #define pm_ctzll(v) ((unsigned) __builtin_ctzll(v))
+#elif defined(_MSC_VER)
+    #include <intrin.h>
+    static inline unsigned pm_ctzll(uint64_t v) {
+        unsigned long index;
+        _BitScanForward64(&index, v);
+        return (unsigned) index;
+    }
+#else
+    static inline unsigned
+    pm_ctzll(uint64_t v) {
+        unsigned c = 0;
+        v &= (uint64_t) (-(int64_t) v);
+        if (v & 0x00000000FFFFFFFFULL) c += 0;  else c += 32;
+        if (v & 0x0000FFFF0000FFFFULL) c += 0;  else c += 16;
+        if (v & 0x00FF00FF00FF00FFULL) c += 0;  else c += 8;
+        if (v & 0x0F0F0F0F0F0F0F0FULL) c += 0;  else c += 4;
+        if (v & 0x3333333333333333ULL) c += 0;  else c += 2;
+        if (v & 0x5555555555555555ULL) c += 0;  else c += 1;
+        return c;
+    }
+#endif
+
 /**
  * We use -Wimplicit-fallthrough to guard potentially unintended fall-through between cases of a switch.
  * Use PRISM_FALLTHROUGH to explicitly annotate cases where the fallthrough is intentional.

prism/node.h

@@ -17,14 +17,31 @@
 #define PM_NODE_LIST_FOREACH(list, index, node) \
     for (size_t index = 0; index < (list)->size && ((node) = (list)->nodes[index]); index++)
 
+/**
+ * Slow path for pm_node_list_append: grow the list and append the node.
+ * Do not call directly — use pm_node_list_append instead.
+ *
+ * @param arena The arena to allocate from.
+ * @param list The list to append to.
+ * @param node The node to append.
+ */
+void pm_node_list_append_slow(pm_arena_t *arena, pm_node_list_t *list, pm_node_t *node);
+
 /**
  * Append a new node onto the end of the node list.
  *
  * @param arena The arena to allocate from.
  * @param list The list to append to.
  * @param node The node to append.
  */
-void pm_node_list_append(pm_arena_t *arena, pm_node_list_t *list, pm_node_t *node);
+static PRISM_FORCE_INLINE void
+pm_node_list_append(pm_arena_t *arena, pm_node_list_t *list, pm_node_t *node) {
+    if (list->size < list->capacity) {
+        list->nodes[list->size++] = node;
+    } else {
+        pm_node_list_append_slow(arena, list, node);
+    }
+}
 
 /**
  * Prepend a new node onto the beginning of the node list.