yeptris 0.6.14.1 → 0.6.15.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 60e8ec13d8221947494e0f163b57a619a3bb6d4f179f53d3f9b324e0ae0a4ffb
-  data.tar.gz: 6eda638121e069206be5d900cc044c942b84e74989d6482255b0e9177aab13f4
+  metadata.gz: 0033737c829c7beb3ae83ca816033386b4fe23b53968f2b0ff40a74c97aac469
+  data.tar.gz: c27accdae4e4807500d4edbfa39228b003c70e5380de3f259bd5315dc9a2dc6f
 SHA512:
-  metadata.gz: dbe32b83a5b017191c37a5978325516a5f99a7d85819ead1dcd97c030fec14f586acfb6d5965275cf1416c74f4dbeecb08abea77074ce769f51340ef3c1e9a7a
-  data.tar.gz: 9b6db9635acbdc9e83017626594b8417fec8c2953e38d17878bf2e99e5a4e0976f4087862890f78a23155ea3eb34d8ca8afeb87e536991fa80091b626e40875e
+  metadata.gz: ca7d477d959df73f231e443f15ceef0403774129e2df0af49d94e2f926d24c380cbacf6e92d88e9488715c028de0533e93fa82ca039ee129dd13b90f4b4badd0
+  data.tar.gz: 54b35dfe9c5399157ffbbb1e33b058afbe523e36222771d000ff697ccaccd059d5b049a3383ac2fbad61086d4548d3bc23343f600311c4c8438535d2a87cc2a8

data/lib/yeptris/document.rb

@@ -26,18 +26,23 @@ class Yeptris::Document
   def self.parse(yaml, schema: :core_12, max_depth: 0)
     yaml = Yeptris.read_input(yaml)
     yaml = yaml.to_s
+    status = ::FFI::MemoryPointer.new(:int)
     doc =
       if schema == :core_12 && max_depth.zero?
-        Yeptris::FFI.yeptris_parse(yaml, yaml.bytesize, nil)
+        Yeptris::FFI.yeptris_parse(yaml, yaml.bytesize, status)
       else
         opts = Yeptris::FFI::ParseOptions.new
         opts[:schema] = schema == :compat_11 ? Yeptris::FFI::SCHEMA_11_COMPAT : Yeptris::FFI::SCHEMA_12_CORE
         opts[:max_depth] = max_depth
-        Yeptris::FFI.yeptris_parse_ex(yaml, yaml.bytesize, opts, nil)
+        Yeptris::FFI.yeptris_parse_ex(yaml, yaml.bytesize, opts, status)
       end
     if doc.null?
-      # the status out-param is skipped: the thread-local error
-      # channel carries the failure detail (measurable on small docs)
+      # NULL + OK is the legal empty stream (no documents): nil, like
+      # stdlib. The status out-param is the discriminator — the TLS
+      # error message cannot serve (a stale error from an earlier
+      # parse persists on the thread).
+      return nil if status.read_int.zero?
+
       raise Yeptris::ParseError,
             "parse failed: #{Yeptris::FFI.last_error_message}"
     end

data/lib/yeptris/node.rb

@@ -276,6 +276,31 @@ class Yeptris::Node
     alive { Yeptris::FFI.yeptris_node_id(@c_ptr) }
   end
 
+  # The marshal fast path (TODO.restructure/21; #178): one C call
+  # turns this subtree into Marshal 4.8 bytes — Marshal.load builds
+  # the Ruby objects in C, no per-node FFI. nil when the document
+  # carries constructs the format cannot express (merge keys,
+  # timestamps, tagged revivals): the caller walks instead. Raises
+  # ParseError on engine errors (not on the bail).
+  def marshal_fast
+    return nil unless Yeptris::FFI::MARSHAL
+
+    out_p = ::FFI::MemoryPointer.new(:pointer)
+    olen_p = ::FFI::MemoryPointer.new(:size_t)
+    st = alive { Yeptris::FFI.yeptris_marshal_node(@c_ptr, out_p, olen_p) }
+    if st == Yeptris::FFI::ERROR_UNSUPPORTED
+      nil
+    elsif st != Yeptris::FFI::OK
+      raise Yeptris::ParseError, Yeptris::FFI.last_error_message
+    else
+      bytes = out_p.read_pointer.read_bytes(olen_p.read_uint64)
+      bytes.force_encoding(Encoding::ASCII_8BIT)
+      ::Marshal.load(bytes)
+    end
+  ensure
+    Yeptris::FFI.yeptris_marshal_free(out_p.read_pointer) if out_p
+  end
+
   def to_ruby(memo = nil)
     # readonly documents memoize per node: a second materialization
     # returns the SAME object (the readonly cache is the ground truth
@@ -293,25 +318,7 @@ class Yeptris::Node
     # Falls back to the per-node FFI walk on constructs the format
     # cannot express (merge keys, timestamps) and on older builds.
     if Yeptris::FFI::MARSHAL && memo.nil?
-      result =
-        begin
-          out_p = ::FFI::MemoryPointer.new(:pointer)
-          olen_p = ::FFI::MemoryPointer.new(:size_t)
-          st = Yeptris::FFI.yeptris_marshal_node(@c_ptr, out_p, olen_p)
-          if st == Yeptris::FFI::ERROR_UNSUPPORTED
-            nil
-          elsif st != Yeptris::FFI::OK
-            raise Yeptris::ParseError, Yeptris::FFI.last_error_message
-          else
-            buf = out_p.read_pointer
-            len = olen_p.read_uint64
-            bytes = buf.read_bytes(len)
-            bytes.force_encoding(Encoding::ASCII_8BIT)
-            ::Marshal.load(bytes)
-          end
-        ensure
-          Yeptris::FFI.yeptris_marshal_free(out_p.read_pointer) if out_p
-        end
+      result = marshal_fast
       unless result.nil?
         @document.readonly_memo[node_id] = result if @document.readonly?
         return result
@@ -332,7 +339,32 @@ class Yeptris::Node
       h = {}
       memo[node_id] = h
       each_pair do |k, v|
-        h[k.to_ruby_key] = v.to_ruby(memo)
+        key = k.to_ruby_key
+        val = v.to_ruby(memo)
+        if key == "<<" && k.tag != "tag:yaml.org,2002:str"
+          # stdlib revive_hash's merge branch: the VALUE's node kind
+          # picks the arm; every merge is TypeError-guarded (a bad
+          # element keeps the whole '<<' pair literal)
+          if v.kind == :alias || v.mapping?
+            begin
+              h.merge!(val)
+            rescue ::TypeError
+              h[key] = val
+            end
+          elsif v.sequence?
+            begin
+              merged = {}
+              val.reverse_each { |e| merged.merge!(e) }
+              h.merge!(merged)
+            rescue ::TypeError
+              h[key] = val
+            end
+          else
+            h[key] = val
+          end
+        else
+          h[key] = val
+        end
       end
       h
     when :sequence
@@ -342,7 +374,11 @@ class Yeptris::Node
       a
     when :alias
       t = alias_target
-      t.nil? ? nil : t.to_ruby(memo)
+      if t.nil?
+        raise ::Yeptris::Psych::AnchorNotDefined,
+              "Unknown anchor: #{anchor}"
+      end
+      t.to_ruby(memo)
     else
       scalar_to_ruby
     end

data/lib/yeptris/psych/visitors.rb

@@ -188,6 +188,13 @@ module Yeptris
               key_node = @tree.new_scalar("", :single_quoted)
               key_node.set_tag("!")
               m.map_add_node(key_node, visit(v))
+            elsif k.is_a?(::String) && k == "<<"
+              # stdlib yaml_tree: a literal chevron key carries the
+              # explicit !!str tag (single-quoted) so it does not
+              # re-load as a merge
+              key_node = @tree.new_scalar("<<", :single_quoted)
+              key_node.set_tag("!!str")
+              m.map_add_node(key_node, visit(v))
             else
               m.map_add(key_text(k), visit(v))
             end
@@ -415,10 +422,38 @@ module Yeptris
           end
         end
 
+        # stdlib visit_hash's merge branch, verbatim in structure: the
+        # key's SHAPE decides (an explicit !!str '<<' is a literal key,
+        # not a merge), the VALUE's node kind picks the arm, and every
+        # merge is TypeError-guarded (a bad element keeps the whole
+        # '<<' pair literal — merges are all-or-nothing).
         def hash_into(h, node)
           anchors[node.anchor] = h if node.anchor
           node.children.each_slice(2) do |k, v|
-            h[visit(k)] = visit(v)
+            key = visit(k)
+            val = visit(v)
+            if key == "<<" && k.tag != "tag:yaml.org,2002:str"
+              case v
+              when ::Yeptris::Psych::Nodes::Alias, ::Yeptris::Psych::Nodes::Mapping
+                begin
+                  h.merge!(val)
+                rescue ::TypeError
+                  h[key] = val
+                end
+              when ::Yeptris::Psych::Nodes::Sequence
+                begin
+                  merged = {}
+                  val.reverse_each { |value| merged.merge!(value) }
+                  h.merge!(merged)
+                rescue ::TypeError
+                  h[key] = val
+                end
+              else
+                h[key] = val
+              end
+            else
+              h[key] = val
+            end
           end
           h
         end

data/lib/yeptris/psych.rb

@@ -91,6 +91,9 @@ module Yeptris
     # object instance exists.
     autoload :Encodable, "yeptris/psych/encodable"
     class Error < StandardError; end
+
+    # stdlib psych 5: a *foo with no matching &foo anchor
+    class AnchorNotDefined < Error; end
     # Psych's exact interface (issue #32): same constructor arity,
     # same reader set (file/line/column/offset/problem/context), same
     # message shape — drop-in consumers' rescues and constructors
@@ -143,6 +146,31 @@ module Yeptris
       # !ruby/set, encode_with/init_with, alias identity. Plain-data
       # loads should use load/safe_load (the Materializer fast path).
       def unsafe_load(yaml, **)
+        # #178: the marshal fast path. Nodes::Builder's tree is per-node
+        # FFI (kind/tag/anchor/value + a walk per container) — the whole
+        # 11 MB relaton index paid ~9.6 s there. When the document is
+        # plain data (no timestamps/merge keys/tagged revivals), ONE C
+        # call materializes the tree as Marshal bytes and Marshal.load
+        # builds the objects in C (~1 s end to end). Construct-heavy
+        # documents keep the full revival walk below, byte for byte.
+        if Yeptris::FFI::MARSHAL && ::Yeptris::Psych.domain_types.empty?
+          begin
+            doc = ::Yeptris::Document.parse(yaml, schema: :compat_11)
+            return nil if doc.nil? || doc.document_count.zero?
+
+            root = doc.root(0)
+            fast = root&.marshal_fast
+            unless fast.nil?
+              doc.free
+              return fast
+            end
+            tree = Nodes::Builder.document(doc) # ownership: the tree frees
+            return Visitors::ToRuby.visit(tree.children.first)
+          rescue ::Yeptris::ParseError => e
+            doc&.free unless doc&.freed?
+            translate_parse_error(e)
+          end
+        end
         tree = parse(yaml)
         return nil if tree.nil?
 
@@ -150,11 +178,16 @@ module Yeptris
       end
 
       def safe_load(yaml, permitted_classes: [::Date, ::Time], aliases: false, **)
-        doc = Yeptris::Document.parse(yaml, schema: :compat_11)
         begin
+          doc = Yeptris::Document.parse(yaml, schema: :compat_11)
+          return nil if doc.nil? # the legal empty stream
+
           force_utf8_scalars(walk_safe(doc.root(0), permitted_classes, aliases))
+        rescue ::Yeptris::ParseError => e
+          doc&.free unless doc&.freed?
+          translate_parse_error(e)
         ensure
-          doc.free
+          doc&.free
         end
       end
 
@@ -230,6 +263,8 @@ module Yeptris
         # children share one C document, so their handles would all
         # resolve to the first document's tree
         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 }
         ensure
@@ -238,9 +273,18 @@ module Yeptris
       end
 
       # The first document's node tree (no Ruby materialization).
+      def translate_parse_error(e)
+        # yeptris rejects *foo without &foo at PARSE time (stdlib
+        # raises at visit) — surface it as stdlib's class so
+        # consumers' rescues keep working
+        raise AnchorNotDefined, e.message if e.message.include?("undefined anchor")
+
+        raise SyntaxError.from_parse_error(e)
+      end
+
       def parse(yaml)
         doc = Yeptris::Document.parse(yaml, schema: :compat_11)
-        return nil if doc.document_count.zero?
+        return nil if doc.nil? || doc.document_count.zero?
 
         Nodes::Builder.document(doc)
       rescue Yeptris::ParseError => e
@@ -249,7 +293,7 @@ module Yeptris
 
       def parse_stream(yaml)
         doc = Yeptris::Document.parse(yaml, schema: :compat_11)
-        return nil if doc.document_count.zero?
+        return nil if doc.nil? || doc.document_count.zero?
 
         stream = Nodes::Stream.new
         (0...doc.document_count).each do |i|

data/lib/yeptris/yaml/descriptor.rb

@@ -33,6 +33,8 @@ class Yeptris::YAML::Descriptor < ::Yeptris::JSON::Descriptor
   def walk(yaml)
     src = ::Yeptris.read_input(yaml).to_s
     doc = ::Yeptris::Document.parse(src, schema: @schema)
+    return [] if doc.nil? # the legal empty stream
+
     begin
       st = ::FFI::MemoryPointer.new(:int)
       raw = ::Yeptris::FFI.yeptris_document_plan_walk(doc.c_ptr, @handle, st)

data/lib/yeptris/yaml.rb

@@ -73,7 +73,7 @@ module Yeptris
     # Parses without materializing: the first document's root Node.
     def parse(yaml, schema: :core_12)
       doc = Document.parse(yaml, schema: schema)
-      return nil if doc.document_count.zero?
+      return nil if doc.nil? || doc.document_count.zero?
 
       doc.root(0)
     end
@@ -177,6 +177,14 @@ module Yeptris
                 off[0] += 1
               elsif k.is_a?(Symbol)
                 scalar(":#{k}", STYLE_PLAIN, emit, blob, off)
+              elsif k.is_a?(String) && k == "<<"
+                # stdlib yaml_tree: a literal chevron key carries the
+                # explicit !!str tag (single-quoted) so it does not
+                # re-load as a merge
+                scalar("<<", STYLE_SQ, emit, blob, off)
+                emit.call(BUILD_TAG, 0, off[0], 5)
+                blob << "!!str"
+                off[0] += 5
               elsif k.is_a?(String)
                 place(k, emit, blob, off, seen)
               else
@@ -343,6 +351,12 @@ module Yeptris
               key_node = doc.new_scalar("", :single_quoted)
               key_node.set_tag("!")
               m.map_add_node(key_node, build(doc, v, seen))
+            elsif k.is_a?(::String) && k == "<<"
+              # stdlib yaml_tree: a literal chevron key carries the
+              # explicit !!str tag so it does not re-load as a merge
+              key_node = doc.new_scalar("<<", :single_quoted)
+              key_node.set_tag("!!str")
+              m.map_add_node(key_node, build(doc, v, seen))
             else
               m.map_add(key_text(k), build(doc, v, seen))
             end

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 = "0.6.14.1".freeze
+  VERSION = "0.6.15.1".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

data/vendor/libyeptris/CMakeLists.txt

@@ -3,7 +3,7 @@
 cmake_minimum_required(VERSION 3.20)
 
 project(yeptris
-    VERSION 0.6.14
+    VERSION 0.6.15
     DESCRIPTION "Ultra-fast YAML 1.2 parser, emitter and streamer in C"
     LANGUAGES C CXX
 )

data/vendor/libyeptris/src/yeptris/marshal.c

@@ -721,6 +721,34 @@ YEPTRIS_API YeptrisStatus yeptris_marshal(const char* data, size_t len, YeptrisS
     return st;
 }
 
+/* #178 (ruby #168's fast path): the value records carry the resolver's
+ * verdict, NOT the source's explicit tag — a `!ruby/object` mapping
+ * would marshal as a plain hash, silently dropping the tag (psych's
+ * visitor revives it). One cheap pre-walk: any explicitly tagged node
+ * in the subtree → UNSUPPORTED, the host walks. Explicit CORE tags
+ * (!!str and friends) ride the same bail — correct, just conservative;
+ * bulk data is untagged. */
+static int dom_subtree_tagged(const yep_dom* d, uint32_t id) {
+    const yep_dnode* n = yep_dom_node(d, id);
+    if (n == NULL) {
+        return 0;
+    }
+    if (n->tag.len != 0) {
+        return 1;
+    }
+    for (uint32_t c = n->first_child; c != UINT32_MAX;) {
+        const yep_dnode* cn = yep_dom_node(d, c);
+        if (cn == NULL) {
+            break;
+        }
+        if (dom_subtree_tagged(d, c) != 0) {
+            return 1;
+        }
+        c = cn->next_sibling;
+    }
+    return 0;
+}
+
 YEPTRIS_API YeptrisStatus yeptris_marshal_node(YeptrisNode node, char** out, size_t* out_len) {
     if (node == NULL || out == NULL || out_len == NULL) {
         return YEPTRIS_ERROR_ARG;
@@ -728,6 +756,12 @@ YEPTRIS_API YeptrisStatus yeptris_marshal_node(YeptrisNode node, char** out, siz
     *out = NULL;
     *out_len = 0;
     yeptris_node* h = (yeptris_node*)node;
+    if (dom_subtree_tagged(h->doc->dom, h->id) != 0) {
+        yep_error_set(yep_error_tls(), YEP_ERR_UNEXPECTED, 0, 0, 0,
+                      "marshal: explicitly tagged node not expressible; "
+                      "fall back to the value walk");
+        return YEPTRIS_ERROR_UNSUPPORTED;
+    }
     yep_value_ctx* c = NULL;
     if (yep_values_from_dom(h->doc->dom, h->id, 0, &c) != 0) {
         return YEPTRIS_ERROR_MEMORY;

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: yeptris
 version: !ruby/object:Gem::Version
-  version: 0.6.14.1
+  version: 0.6.15.1
 platform: ruby
 authors:
 - Ribose Inc.