yeptris 0.6.16.1-aarch64-linux → 0.6.16.2-aarch64-linux

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e2612a8d5390e6bf215a646fee2c582ff4aaa1091a65d26fb0f113220983741c
-  data.tar.gz: c706b95796ded1c06c4ea4a41415ee484c6f894d9fdbf8652f20d3f3adae5c9b
+  metadata.gz: 9505473e51a840dbf7c50d8ed7ae56b7e3a9b3592e26787ba0a0b231320bfe86
+  data.tar.gz: 78fbfa73c88cc29e5376cde241e310b1b4d65b878e96198faf733640bc695454
 SHA512:
-  metadata.gz: 0ad96188aa09042b35024144cdb9b555abfc0bf5731d73f4d246b4d84cbd2b001080f769d3fa8a0f1488b3ce41da672881724fb21acb431a74ddbb43a468664f
-  data.tar.gz: 496b1c54f30d326cf2999fca95105c56756bf4addb32b32e4d46651a2930921973bda3c8bf488e23b7c63185be3223210cc1deeb2e2e9a1fc2dfd39e5163c185
+  metadata.gz: 88c4aaa0e2b48a7b5ce25da76e0035e0cd2d83f2901bde878c6114cdf046ade9c9e7ae74ff70ebc3fe403ea98a5dce86504de060cec35a5fb2e4e6a58a4c83b6
+  data.tar.gz: ed0b241f30ae93ea79f8d4deae9ca1f33dd14c124698484c79ea9905e42149564487bbad18234f704125185b0583f7a740c76817eaa2c522b28fa2f3dce20240

data/ext/yeptris_native/yeptris_native.c

@@ -423,8 +423,51 @@ static VALUE native_gc_mode_set(VALUE self, VALUE mode) {
     return mode;
 }
 
+/* ABI-drift self-check (the #157-audit silent-nil class): the bundle
+ * is compiled against a specific yep_dom layout; engine drift makes
+ * the walkers read fields at stale offsets — cbor_load returned NIL
+ * on valid CBOR in the dev-checkout audit, with no error anywhere.
+ * Both walkers must round-trip a known document at load, or Init
+ * raises LoadError: lib/yeptris.rb's existing rescue falls back to
+ * the FFI ladder (loudly), and the stale bundle never answers with
+ * garbage. Runs inside rb_protect so a drifted walk raises instead of
+ * crashing the load. */
+static VALUE native_self_check_body(VALUE unused) {
+    /* JSON: {"a"=>[1, 2.5, "x", true, nil]} */
+    const char* json = "{\"a\":[1,2.5,\"x\",true,null]}";
+    VALUE j = yep_rb_parse_json(json, strlen(json), 0);
+    if (j == Qundef || !RB_TYPE_P(j, T_HASH)) return Qfalse;
+    VALUE a = rb_hash_lookup(j, rb_str_new_cstr("a"));
+    if (!RB_TYPE_P(a, T_ARRAY) || RARRAY_LEN(a) != 5) return Qfalse;
+    if (!RB_INTEGER_TYPE_P(rb_ary_entry(a, 0))) return Qfalse;
+    if (!RB_FLOAT_TYPE_P(rb_ary_entry(a, 1))) return Qfalse;
+    if (!RB_TYPE_P(rb_ary_entry(a, 2), T_STRING)) return Qfalse;
+    if (rb_ary_entry(a, 3) != Qtrue || !NIL_P(rb_ary_entry(a, 4))) return Qfalse;
+
+    /* CBOR: A1 61 6B 01 = {"k" => 1} (canonical, hand-encoded) */
+    const char cbor[] = "\xA1\x61\x6B\x01";
+    VALUE c = yep_rb_cbor_load(cbor, sizeof(cbor) - 1, 0);
+    if (c == Qundef || !RB_TYPE_P(c, T_HASH)) return Qfalse;
+    VALUE one = rb_hash_lookup(c, rb_str_new_cstr("k"));
+    if (!RB_INTEGER_TYPE_P(one) || !rb_eql(one, INT2FIX(1))) return Qfalse;
+    return Qtrue;
+}
+
+static void native_self_check(void) {
+    int state = 0;
+    VALUE ok = rb_protect(native_self_check_body, Qnil, &state);
+    if (state != 0 || ok != Qtrue) {
+        if (state != 0) rb_set_errinfo(Qnil);
+        rb_raise(rb_eLoadError,
+                 "yeptris: native materializer failed its ABI self-check "
+                 "(stale build against a drifted engine?) — refusing to "
+                 "register; the FFI ladder will carry the load");
+    }
+}
+
 RUBY_FUNC_EXPORTED void Init_native(void) {
     utf8_enc = rb_utf8_encoding();
+    native_self_check();
     VALUE mYep = rb_define_module("Yeptris");
     VALUE mNat = rb_define_module_under(mYep, "Native");
     rb_define_singleton_method(mNat, "load_json", native_load_json, -1);

data/lib/yeptris/json.rb

@@ -119,7 +119,7 @@ module Yeptris
       end
     end
 
-    def walk_tape(src, tape)
+    def walk_tape(src, tape, strict_dup = STRICT_DUPLICATE_KEYS)
       n = tape[:count]
       kinds = tape[:kinds].read_bytes(n).unpack("C*")
       offs = tape[:offs].read_bytes(n * 4).unpack("V*")
@@ -139,7 +139,7 @@ module Yeptris
 
       docs = [nil] # record 0 (DOC) pre-consumed — the root's slot
       stack = []
-      key_sets = STRICT_DUPLICATE_KEYS ? [{}] : nil
+      key_sets = strict_dup ? [{}] : nil
       pending_key = nil
       i = 1
       while i < n
@@ -152,8 +152,12 @@ module Yeptris
           place(docs, stack, pending_key) { {} }
           pending_key = nil
         when T_CLOSE
+          # Key frames are pushed per MAP_OPEN — an ARRAY close must
+          # not pop the enclosing map's frame (issue found by ea's CI:
+          # {"a":[1],"b":[2],"c":[3]} exhausted the frames and the next
+          # map key hit key_sets.last.key? on nil under strict mode).
+          key_sets&.pop if stack.last.is_a?(Hash)
           stack.pop
-          key_sets&.pop
         when T_STR
           # empty-string literals are frozen+shared since Ruby 3.4
           # (and `+""` binds after the method chain anyway), so the
@@ -256,8 +260,8 @@ module Yeptris
           place(docs, stack, pending_key) { {} }
           pending_key = nil
         when ValueML::CLOSE
+          key_sets&.pop if stack.last.is_a?(Hash)
           stack.pop
-          key_sets&.pop
         when ValueML::V_STR
           text = arena.byteslice(offs[i], lens[i])
           if ikeys[i] == 1 && !stack.empty? && stack.last.is_a?(Hash)

data/lib/yeptris/psych/class_loader.rb

@@ -0,0 +1,24 @@
+# frozen_string_literal: true
+
+require "date"
+
+module Yeptris
+  module Psych
+    # The minimal ClassLoader face the ScalarScanner port drives
+    # (stdlib's loader resolves through the class hierarchy; the
+    # scanner only needs the date class and symbolize).
+    class ClassLoader
+      def date
+        Date
+      end
+
+      def symbolize str
+        str.to_sym
+      end
+
+      def load name
+        Object.const_get(name)
+      end
+    end
+  end
+end

data/lib/yeptris/psych/scalar_scanner.rb

@@ -0,0 +1,149 @@
+# frozen_string_literal: true
+
+require "date"
+
+module Yeptris
+  module Psych
+    ###
+    # Scan scalars for built in types — the stdlib Psych::ScalarScanner
+    # port (#179): same regexes, same precedence, same edge verdicts.
+    # The load path types through the C resolver; this class is the
+    # standalone API stdlib code (and Psych's own tests) drive directly.
+    class ScalarScanner
+      # Taken from http://yaml.org/type/timestamp.html
+      TIME = /^-?\d{4}-\d{1,2}-\d{1,2}(?:[Tt]|\s+)\d{1,2}:\d\d:\d\d(?:\.\d*)?(?:\s*(?:Z|[-+]\d{1,2}:?(?:\d\d)?))?$/
+
+      # Taken from http://yaml.org/type/float.html
+      # Base 60, [-+]inf and NaN are handled separately
+      FLOAT = /^(?:[-+]?([0-9][0-9_,]*)?\.[0-9]*([eE][-+][0-9]+)?(?# base 10))$/x
+
+      # Taken from http://yaml.org/type/int.html and modified to ensure at least one numerical symbol exists
+      INTEGER_STRICT = /^(?:[-+]?0b[_]*[0-1][0-1_]*             (?# base 2)
+                           |[-+]?0[_]*[0-7][0-7_]*              (?# base 8)
+                           |[-+]?(0|[1-9][0-9_]*)               (?# base 10)
+                           |[-+]?0x[_]*[0-9a-fA-F][0-9a-fA-F_]* (?# base 16))$/x
+
+      # Same as above, but allows commas.
+      # Not to YML spec, but kept for backwards compatibility
+      INTEGER_LEGACY = /^(?:[-+]?0b[_,]*[0-1][0-1_,]*                (?# base 2)
+                           |[-+]?0[_,]*[0-7][0-7_,]*                 (?# base 8)
+                           |[-+]?(?:0|[1-9](?:[0-9]|,[0-9]|_[0-9])*) (?# base 10)
+                           |[-+]?0x[_,]*[0-9a-fA-F][0-9a-fA-F_,]*    (?# base 16))$/x
+
+      BOOLEAN_TRUE  = /^(yes|true|on)$/i
+      BOOLEAN_FALSE = /^(no|false|off)$/i
+
+      attr_reader :class_loader
+
+      # Create a new scanner
+      def initialize class_loader = ClassLoader, strict_integer: false, parse_symbols: true
+        @symbol_cache = {}
+        @class_loader = class_loader
+        @strict_integer = strict_integer
+        @parse_symbols = parse_symbols
+      end
+
+      # Tokenize +string+ returning the Ruby object
+      def tokenize string
+        return nil if string.empty?
+        return @symbol_cache[string] if @symbol_cache.key?(string)
+        integer_regex = @strict_integer ? INTEGER_STRICT : INTEGER_LEGACY
+        # Check for a String type, being careful not to get caught by hash keys, hex values, and
+        # special floats (e.g., -.inf).
+        if string.match?(%r{^[^\d.:-]?[[:alpha:]_\s!@#$%\^&*(){}<>|/\\~;=]+}) || string.match?(/\n/)
+          return string if string.length > 5
+
+          if string.match?(/^[^ytonf~]/i)
+            string
+          elsif string == '~' || string.match?(/^null$/i)
+            nil
+          elsif string.match?(BOOLEAN_TRUE)
+            true
+          elsif string.match?(BOOLEAN_FALSE)
+            false
+          else
+            string
+          end
+        elsif string.match?(TIME)
+          begin
+            parse_time string
+          rescue ArgumentError
+            string
+          end
+        elsif string.match?(/^\d{4}-(?:1[012]|0\d|\d)-(?:[12]\d|3[01]|0\d|\d)$/)
+          begin
+            class_loader.date.strptime(string, '%F', Date::GREGORIAN)
+          rescue ArgumentError
+            string
+          end
+        elsif string.match?(/^\+?\.inf$/i)
+          Float::INFINITY
+        elsif string.match?(/^-\.inf$/i)
+          -Float::INFINITY
+        elsif string.match?(/^\.nan$/i)
+          Float::NAN
+        elsif @parse_symbols && string.match?(/^:./)
+          if string =~ /^:(["'])(.*)\1/
+            @symbol_cache[string] = class_loader.symbolize($2.sub(/^:/, ''))
+          else
+            @symbol_cache[string] = class_loader.symbolize(string.sub(/^:/, ''))
+          end
+        elsif string.match?(/^[-+]?[0-9][0-9_]*(:[0-5]?[0-9]){1,2}$/)
+          i = 0
+          string.split(':').each_with_index do |n, e|
+            i += (n.to_i * 60**(e - 2).abs)
+          end
+          i
+        elsif string.match?(/^[-+]?[0-9][0-9_]*(:[0-5]?[0-9]){1,2}\.[0-9_]*$/)
+          i = 0
+          string.split(':').each_with_index do |n, e|
+            i += (n.to_f * 60**(e - 2).abs)
+          end
+          i
+        elsif string.match?(FLOAT)
+          if string.match?(/\A[-+]?\.\Z/)
+            string
+          else
+            Float(string.delete(',_').gsub(/\.([Ee]|$)/, '\1'))
+          end
+        elsif string.match?(integer_regex)
+          parse_int string
+        else
+          string
+        end
+      end
+
+      # Parse and return an int from +string+
+      def parse_int string
+        Integer(string.delete(',_'))
+      end
+
+      ###
+      # Parse and return a Time from +string+
+      def parse_time string
+        date, time = *(string.split(/[Tt]|\s+/, 2))
+        (yy, m, dd) = date.match(/^(-?\d{4})-(\d{1,2})-(\d{1,2})/).captures.map { |x| x.to_i }
+        md = time.match(/(\d+:\d+:\d+)(?:\.(\d*))?\s*(Z|[-+]\d+(:\d\d)?)?/)
+
+        (hh, mm, ss) = md[1].split(':').map { |x| x.to_i }
+        us = (md[2] ? Rational("0.#{md[2]}") : 0) * 1_000_000
+
+        time = Time.utc(yy, m, dd, hh, mm, ss, us)
+
+        return time if 'Z' == md[3]
+        return Time.at(time.to_i, us) unless md[3]
+
+        tz = md[3].match(/^([+\-]?\d{1,2})\:?(\d{1,2})?$/)[1..-1].compact.map { |digit| Integer(digit, 10) }
+        offset = tz.first * 3600
+
+        if offset < 0
+          offset -= ((tz[1] || 0) * 60)
+        else
+          offset += ((tz[1] || 0) * 60)
+        end
+
+        Time.new(yy, m, dd, hh, mm, ss + us / 1_000_000r, offset)
+      end
+    end
+  end
+end

data/lib/yeptris/psych.rb

@@ -22,6 +22,9 @@ require "yeptris"
 # document without materializing.
 module Yeptris
   module Psych
+    autoload :ClassLoader, "yeptris/psych/class_loader"
+    autoload :ScalarScanner, "yeptris/psych/scalar_scanner"
+
     # The tag registries (Psych's class-level API, #95 bug 4):
     # load_tags maps a serialized tag to the Class that revives it;
     # dump_tags overrides the emitted tag for a Class. Consulted by
@@ -258,18 +261,28 @@ module Yeptris
         ::Yeptris::Psych.dump(obj, io, options)
       end
 
-      def load_stream(yaml, **kwargs)
+      def load_stream(yaml, **kwargs, &block)
         # materialize each document's root directly — the stream
         # children share one C document, so their handles would all
-        # resolve to the first document's tree
+        # resolve to the first document's tree. stdlib's block form
+        # yields each loaded document's object (#179 round 4).
         doc = Yeptris::Document.parse(yaml, schema: :compat_11)
         return nil if doc.nil? # the legal empty stream
 
         begin
-          (0...doc.document_count).map { |i| doc.root(i).to_ruby }
+          docs = (0...doc.document_count).map { |i| doc.root(i).to_ruby }
         ensure
           doc.free
         end
+        docs.each { |d| block.call(d) } if block
+        docs
+      end
+
+      # safe_load_stream: stdlib's surface — the stream form of
+      # safe_load. Our load_stream is already safe-by-default
+      # (Psych 5 semantics), so this is the yielding wrapper.
+      def safe_load_stream(yaml, **kwargs, &block)
+        load_stream(yaml, **kwargs, &block)
       end
 
       # The first document's node tree (no Ruby materialization).
@@ -299,13 +312,14 @@ module Yeptris
         # handles GC-free. (The block-yielding form is tracked under
         # #179: the current implementation materializes eagerly; the
         # block is only honored as a no-op convenience.)
-        _ = block # reserved for the future yielding form (#179)
         doc = Yeptris::Document.parse(yaml, schema: :compat_11)
         return nil if doc.nil? || doc.document_count.zero?
 
         stream = Nodes::Stream.new
         (0...doc.document_count).each do |i|
-          stream.children << Nodes::Builder.document_stream_child(doc, i)
+          child = Nodes::Builder.document_stream_child(doc, i)
+          stream.children << child
+          block.call(child) if block # stdlib's yielding form (#179 round 4)
         end
         # ownership: the DOCUMENT wrapper is the sole owner — its own
         # finalizer (pointer-only closure) frees the C memory; the
@@ -450,6 +464,10 @@ module Yeptris
       # tree stay valid while the tree is reachable; a GC finalizer
       # releases the C memory when it is not.
       class Document < Node
+          # The document's root as a Ruby object (the stream-yield face).
+          def to_ruby
+            children.first&.to_ruby
+          end
         attr_reader :version, :tags
 
         def initialize(version = [], tags = {})

data/lib/yeptris.rb

@@ -3,7 +3,7 @@
 module Yeptris
   # The gem's version lives in the parent namespace's file — the last
   # internal require (yeptris/version) retired with it.
-  VERSION = "0.6.16.1".freeze
+  VERSION = "0.6.16.2".freeze
   # The error hierarchy lives in THIS file (the parent namespace's
   # own file): nested constants do not trigger a parent-constant
   # autoload, and the law forbids internal requires — defining the

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: yeptris
 version: !ruby/object:Gem::Version
-  version: 0.6.16.1
+  version: 0.6.16.2
 platform: aarch64-linux
 authors:
 - Ribose Inc.
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2026-09-21 00:00:00.000000000 Z
+date: 2026-09-22 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: ffi
@@ -50,11 +50,13 @@ files:
 - lib/yeptris/native-3.3.so
 - lib/yeptris/node.rb
 - lib/yeptris/psych.rb
+- lib/yeptris/psych/class_loader.rb
 - lib/yeptris/psych/coder_shim.rb
 - lib/yeptris/psych/drop_in.rb
 - lib/yeptris/psych/encodable.rb
 - lib/yeptris/psych/handler.rb
 - lib/yeptris/psych/parser.rb
+- lib/yeptris/psych/scalar_scanner.rb
 - lib/yeptris/psych/visitors.rb
 - lib/yeptris/schema.rb
 - lib/yeptris/valueml.rb