Define String#ascii_only? and #valid_encoding? in Ruby

Get rid of the CFUNC overhead in String#ascii_only? and #valid_encoding?. Define them as leaf Primitive.cexpr! builtins in a new string.rb, following the pattern of numeric.rb.

This makes them 1.2x-1.4x faster across the interpreter, YJIT, and
ZJIT.

Move ZJIT's cfunc annotation for String#ascii_only? to annotate_builtin!, and add one for String#valid_encoding?.

Fix CI for builtin String#ascii_only?

Author: sampokuokkanen

benchmark/string_ascii_only_p.yml

@@ -0,0 +1,7 @@
+prelude: |
+  ascii = "hello world"
+  utf8 = "héllo wörld 晦"
+benchmark:
+  ascii_only_true: ascii.ascii_only?
+  ascii_only_false: utf8.ascii_only?
+loop_count: 20000000

benchmark/string_valid_encoding_p.yml

@@ -0,0 +1,7 @@
+prelude: |
+  utf8 = "héllo wörld 晦"
+  broken = "\xff\xfe".dup.force_encoding("UTF-8")
+benchmark:
+  valid_enc_true: utf8.valid_encoding?
+  valid_enc_false: broken.valid_encoding?
+loop_count: 20000000

common.mk

@@ -1199,6 +1199,7 @@ BUILTIN_RB_SRCS = \
 		$(srcdir)/kernel.rb \
 		$(srcdir)/pathname_builtin.rb \
 		$(srcdir)/ractor.rb \
+		$(srcdir)/string.rb \
 		$(srcdir)/symbol.rb \
 		$(srcdir)/timev.rb \
 		$(srcdir)/thread_sync.rb \

depend

@@ -17718,6 +17718,7 @@ string.$(OBJEXT): {$(VPATH)}backward/2/limits.h
 string.$(OBJEXT): {$(VPATH)}backward/2/long_long.h
 string.$(OBJEXT): {$(VPATH)}backward/2/stdalign.h
 string.$(OBJEXT): {$(VPATH)}backward/2/stdarg.h
+string.$(OBJEXT): {$(VPATH)}builtin.h
 string.$(OBJEXT): {$(VPATH)}config.h
 string.$(OBJEXT): {$(VPATH)}constant.h
 string.$(OBJEXT): {$(VPATH)}debug_counter.h
@@ -17896,6 +17897,7 @@ string.$(OBJEXT): {$(VPATH)}rubyparser.h
 string.$(OBJEXT): {$(VPATH)}shape.h
 string.$(OBJEXT): {$(VPATH)}st.h
 string.$(OBJEXT): {$(VPATH)}string.c
+string.$(OBJEXT): {$(VPATH)}string.rbinc
 string.$(OBJEXT): {$(VPATH)}subst.h
 string.$(OBJEXT): {$(VPATH)}thread.h
 string.$(OBJEXT): {$(VPATH)}thread_$(THREAD_MODEL).h

inits.c

@@ -105,6 +105,7 @@ rb_call_builtin_inits(void)
     BUILTIN(warning);
     BUILTIN(array);
     BUILTIN(hash);
+    BUILTIN(string);
     BUILTIN(symbol);
     BUILTIN(timev);
     BUILTIN(thread_sync);

string.c

@@ -11736,14 +11736,6 @@ rb_str_b(VALUE str)
     return str2;
 }
 
-/*
- *  call-seq:
- *    valid_encoding? -> true or false
- *
- *  :include: doc/string/valid_encoding_p.rdoc
- *
- */
-
 static VALUE
 rb_str_valid_encoding_p(VALUE str)
 {
@@ -11752,18 +11744,6 @@ rb_str_valid_encoding_p(VALUE str)
     return RBOOL(cr != ENC_CODERANGE_BROKEN);
 }
 
-/*
- *  call-seq:
- *    ascii_only? -> true or false
- *
- *  Returns whether +self+ contains only ASCII characters:
- *
- *    'abc'.ascii_only?         # => true
- *    "abc\u{6666}".ascii_only? # => false
- *
- *  Related: see {Querying}[rdoc-ref:String@Querying].
- */
-
 static VALUE
 rb_str_is_ascii_only_p(VALUE str)
 {
@@ -13053,8 +13033,6 @@ Init_String(void)
     rb_define_method(rb_cString, "encoding", rb_obj_encoding, 0); /* in encoding.c */
     rb_define_method(rb_cString, "force_encoding", rb_str_force_encoding, 1);
     rb_define_method(rb_cString, "b", rb_str_b, 0);
-    rb_define_method(rb_cString, "valid_encoding?", rb_str_valid_encoding_p, 0);
-    rb_define_method(rb_cString, "ascii_only?", rb_str_is_ascii_only_p, 0);
 
     /* define UnicodeNormalize module here so that we don't have to look it up */
     mUnicodeNormalize          = rb_define_module("UnicodeNormalize");
@@ -13107,3 +13085,5 @@ Init_String(void)
     rb_define_method(rb_cSymbol, "encoding", sym_encoding, 0);
 }
 
+#include "string.rbinc"
+

string.rb

@@ -0,0 +1,25 @@
+class String
+  #  call-seq:
+  #    valid_encoding? -> true or false
+  #
+  #  :include: doc/string/valid_encoding_p.rdoc
+  #
+  def valid_encoding?
+    Primitive.attr! :leaf
+    Primitive.cexpr! 'rb_str_valid_encoding_p(self)'
+  end
+
+  #  call-seq:
+  #    ascii_only? -> true or false
+  #
+  #  Returns whether +self+ contains only ASCII characters:
+  #
+  #    'abc'.ascii_only?         # => true
+  #    "abc\u{6666}".ascii_only? # => false
+  #
+  #  Related: see {Querying}[rdoc-ref:String@Querying].
+  def ascii_only?
+    Primitive.attr! :leaf
+    Primitive.cexpr! 'rb_str_is_ascii_only_p(self)'
+  end
+end

zjit/src/cruby_methods.rs

@@ -222,9 +222,6 @@ pub fn init() -> Annotations {
     annotate!(rb_cString, "<<", inline_string_append);
     annotate!(rb_cString, "==", inline_string_eq);
     // Not elidable; has a side effect of setting the encoding if ENC_CODERANGE_UNKNOWN.
-    // TOOD(max): Turn this into a load/compare. Will need to side-exit or do the full call if
-    // ENC_CODERANGE_UNKNOWN.
-    annotate!(rb_cString, "ascii_only?", types::BoolExact, no_gc, leaf);
     annotate!(rb_cModule, "name", types::StringExact.union(types::NilClass), no_gc, leaf, elidable);
     annotate!(rb_cModule, "===", inline_module_eqq, types::BoolExact, no_gc, leaf);
     annotate!(rb_cArray, "length", inline_array_length, types::Fixnum, no_gc, leaf, elidable);
@@ -291,6 +288,10 @@ pub fn init() -> Annotations {
     annotate_builtin!(rb_mKernel, "frozen?", types::BoolExact);
     annotate_builtin!(rb_cSymbol, "name", types::StringExact);
     annotate_builtin!(rb_cSymbol, "to_s", types::StringExact);
+    // TOOD(max): Turn this into a load/compare. Will need to side-exit or do the full call if
+    // ENC_CODERANGE_UNKNOWN.
+    annotate_builtin!(rb_cString, "ascii_only?", types::BoolExact, no_gc, leaf);
+    annotate_builtin!(rb_cString, "valid_encoding?", types::BoolExact, no_gc, leaf);
 
     // Array iteration builtins (used in with_jit Array#each, map, select, find)
     builtin_funcs.insert(rb_jit_fixnum_inc as *mut c_void, FnProperties { inline: inline_fixnum_inc, return_type: types::Fixnum, ..Default::default() });

zjit/src/hir/opt_tests.rs

@@ -11910,8 +11910,8 @@ mod hir_opt_tests {
         bb3(v9:BasicObject, v10:BasicObject):
           PatchPoint NoSingletonClass(String@0x1008)
           PatchPoint MethodRedefined(String@0x1008, ascii_only?@0x1010, cme:0x1018)
-          v24:StringExact = GuardType v10, StringExact recompile
-          v25:BoolExact = CCall v24, :String#ascii_only?@0x1040
+          v23:StringExact = GuardType v10, StringExact recompile
+          v25:BoolExact = InvokeBuiltin leaf <inline_expr>, v23
           CheckInterrupts
           Return v25
         ");