yeptris 0.6.15.2-aarch64-linux → 0.6.16.1-aarch64-linux

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 419470a0e038bec65880a3d764da46f072d7993e5aca9de71825a8428d5b2f98
-  data.tar.gz: 19cf714c7f80abc07d8729420700ef2c08bf1a140de03c3e11a8bb6ff177a4f4
+  metadata.gz: e2612a8d5390e6bf215a646fee2c582ff4aaa1091a65d26fb0f113220983741c
+  data.tar.gz: c706b95796ded1c06c4ea4a41415ee484c6f894d9fdbf8652f20d3f3adae5c9b
 SHA512:
-  metadata.gz: 9d11c53e5305c76f780f431d73be9c62228e250ab164c4fd636429ad4fca4e3705b1d46b0e9cf7abed5d8a0662d0347a2399de0ce3a16f5eec987d05d0fb053f
-  data.tar.gz: 5c24e95902b4ae44595710274ede3efbd605885d8cbc0747fc0965fcb8a2addb32ea8f18648606fef823d09a6f106d777d175f4e8ed0f1c8e74810f22d6cbee6
+  metadata.gz: 0ad96188aa09042b35024144cdb9b555abfc0bf5731d73f4d246b4d84cbd2b001080f769d3fa8a0f1488b3ce41da672881724fb21acb431a74ddbb43a468664f
+  data.tar.gz: 496b1c54f30d326cf2999fca95105c56756bf4addb32b32e4d46651a2930921973bda3c8bf488e23b7c63185be3223210cc1deeb2e2e9a1fc2dfd39e5163c185

data/ext/yeptris_native/cbor_ruby.c

@@ -17,6 +17,14 @@
 #include <stdlib.h>
 #include <string.h>
 
+static rb_encoding* cr_utf8_enc;
+
+/* Map keys repeat across records in CBOR corpora; interning shares one
+ * VALUE per distinct key (the json_ruby.c on_key pattern, #157). */
+static VALUE cr_key_str(const char* p, size_t len) {
+    return rb_enc_interned_str(p, (long)len, cr_utf8_enc);
+}
+
 #include "dom/dom.h"
 #include "doc.h" /* the public YeptrisDocument wrapper: ->dom */
 #include <yeptris/cbor.h>
@@ -91,8 +99,17 @@ static VALUE cr_walk(const yep_dom* d, uint32_t id) {
         VALUE h = HASH_NEW_CAPA(pairs);
         uint32_t c = n->first_child;
         for (long i = 0; i < pairs && c != UINT32_MAX; i++) {
-            VALUE k = cr_walk(d, c);
-            c = d->nodes[c].next_sibling;
+            VALUE k;
+            const yep_dnode* kn = &d->nodes[c];
+            if (kn->kind == YEP_DOM_SCALAR && kn->tag_id == YEPTRIS_TAG_STR) {
+                uint32_t klen = 0;
+                const char* kp = cr_view(d, kn->value, &klen);
+                k = cr_key_str(kp, klen);
+                c = kn->next_sibling;
+            } else {
+                k = cr_walk(d, c);
+                c = d->nodes[c].next_sibling;
+            }
             VALUE v = (c != UINT32_MAX) ? cr_walk(d, c) : Qnil;
             if (c != UINT32_MAX) {
                 c = d->nodes[c].next_sibling;
@@ -107,6 +124,9 @@ static VALUE cr_walk(const yep_dom* d, uint32_t id) {
 }
 
 VALUE yep_rb_cbor_load(const char* p, size_t len, int strict) {
+    if (cr_utf8_enc == NULL) {
+        cr_utf8_enc = rb_utf8_encoding();
+    }
     YeptrisStatus st = YEPTRIS_OK;
     /* the DOM borrows the input buffer zero-copy; the walk ALLOCATES,
      * and a GC compaction mid-walk moves the caller's String — the

data/lib/yeptris/document.rb

@@ -8,6 +8,11 @@ class Yeptris::Document
   # explicit free path and the finalizer both flip the same flag.
   Freed = Struct.new(:state) # :alive | :freed
 
+  # The raw C handle — the Yeptris::Node wrappers cache by its
+  # address; stream-children wrappers (Document.without_finalizer)
+  # carry it without registering a finalizer (#182).
+  attr_reader :c_ptr
+
   # The schema this document was parsed with (a parse property).
   attr_reader :parse_schema
 
@@ -80,6 +85,18 @@ class Yeptris::Document
     new(c_ptr, Freed.new(:alive), schema)
   end
 
+  # #182: the stream-children wrapper owns NOTHING (the parent
+  # Document wrapper is the sole freer of the C memory). A plain
+  # Document.new would register a finalizer on a c_ptr we don't own;
+  # a parse_stream on a multi-doc stream would free the same pointer
+  # three times at GC → SIGABRT. This factory builds the wrapper
+  # WITHOUT a finalizer; the wrapper's lifetime follows the owner.
+  def self.without_finalizer(c_ptr)
+    doc = new(c_ptr)
+    ObjectSpace.undefine_finalizer(doc)
+    doc
+  end
+
   def ensure_alive!
     raise Yeptris::FreedError, "document is freed" if @freed.state == :freed
   end

data/lib/yeptris/psych.rb

@@ -291,7 +291,15 @@ module Yeptris
         raise SyntaxError.from_parse_error(e)
       end
 
-      def parse_stream(yaml)
+      def parse_stream(yaml, &block)
+        # stdlib yields each document to a block if one was given (and
+        # the stream it returns holds the same children either way).
+        # #182: the wrapper is the sole owner of the C memory — the
+        # stream merely REFERENCES it; the wrapper's own finalizer
+        # 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?
 
@@ -308,6 +316,8 @@ module Yeptris
       rescue Yeptris::ParseError => e
         raise SyntaxError.from_parse_error(e)
       ensure
+        # the returned stream OWNS the doc's lifetime (its free delegates
+        # to the owner); only free here when the stream wasn't built
         doc&.free if doc && !stream
       end
 
@@ -522,8 +532,12 @@ module Yeptris
         # stream owns the yeptris document)
         def document_stream_child(doc, index)
           root = doc.root(index)
-          d = Document.new
-          d.handle = root&.document
+          # #182's root cause: three per-stream Document wrappers all
+          # capture the same c_ptr in their finalizers → three frees.
+          # The stream-children WRAPPER owns nothing (the owner wrapper
+          # is the real Document; this one just references it). Build
+          # a non-owning wrapper with no finalizer.
+          d = Document.send(:new, root&.document, false)
           d.children << node(root) if root
           d
         end

data/lib/yeptris/schema.rb

@@ -66,6 +66,92 @@ module Yeptris
       end
     end
 
+    # #184 (lutaml-model KV path): zip Schema columns into record
+    # hashes ready for Serializable.instantiate. Mapping root → one
+    # hash; sequence-of-mappings → one hash per element; nested
+    # mappings → nested hashes. Returns Array<Hash> always. when_attribute
+    # / polymorphic stay out of scope (interpretive fallback).
+    def load_records(source, schema: :core_12, desc:, capacity: 64)
+      cols = load(source, schema: schema, desc: desc, capacity: capacity)
+      zip_records(desc, cols)
+    end
+
+    def zip_records(desc, cols)
+      root = desc[0] || {}
+      case root[:kind]
+      when :sequence
+        child_start = root[:child_index] || 1
+        child_count = root[:child_count] || 0
+        return [] if child_count.zero?
+
+        child = desc[child_start]
+        if child && child[:kind] == :mapping
+          field_start = child[:child_index] || (child_start + 1)
+          field_count = child[:child_count] || 0
+          fields = desc[field_start, field_count] || []
+          field_cols = cols[field_start, field_count] || []
+          nrows = field_cols.map(&:length).max || 0
+          Array.new(nrows) do |r|
+            h = {}
+            fields.each_with_index do |f, i|
+              next unless f[:wire_name]
+              col = field_cols[i] || []
+              # ABSENT keys are OMITTED, not nil-filled (lutaml-model's
+              # load-bearing edge: absent seeds defaults +
+              # using_default?; explicit nil is a present value with
+              # different render semantics). The C columns fill in
+              # document order — col.length > r means present.
+              next if col.length <= r
+
+              h[f[:wire_name].to_sym] = materialize_field(f, desc, cols, col[r])
+            end
+            h
+          end
+        else
+          (cols[child_start] || []).map { |v| { value: v } }
+        end
+      when :mapping
+        field_start = root[:child_index] || 1
+        field_count = root[:child_count] || 0
+        fields = desc[field_start, field_count] || []
+        field_cols = cols[field_start, field_count] || []
+        h = {}
+        fields.each_with_index do |f, i|
+          next unless f[:wire_name]
+          col = field_cols[i] || []
+          next if col.empty? # absent: omitted, not nil-filled (#184)
+
+          h[f[:wire_name].to_sym] = materialize_field(f, desc, cols, col[0])
+        end
+        [h]
+      else
+        [{ value: cols[0]&.first }]
+      end
+    end
+    private_class_method :zip_records
+
+    def materialize_field(field, desc, cols, cell)
+      case field[:kind]
+      when :mapping
+        start = field[:child_index] || 0
+        count = field[:child_count] || 0
+        return nil if count.zero? || cell.nil?
+
+        nested = {}
+        desc[start, count]&.each_with_index do |nf, i|
+          next unless nf[:wire_name]
+          ncol = cols[start + i] || []
+          nested[nf[:wire_name].to_sym] = ncol.is_a?(Array) ? (ncol[0] rescue cell) : cell
+        end
+        nested
+      when :sequence
+        cell.is_a?(Array) ? cell : Array(cell)
+      else
+        cell
+      end
+    end
+    private_class_method :materialize_field
+
     def compile(desc)
       desc.map do |n|
         { kind: KIND.fetch(n.fetch(:kind)),

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 = "0.6.15.2".freeze
+  VERSION = "0.6.16.1".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 0.6.15
+    VERSION 0.6.16
     DESCRIPTION "Ultra-fast YAML 1.2 parser, emitter and streamer in C"
     LANGUAGES C CXX
 )

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

@@ -241,6 +241,9 @@ void dom_link(yep_dom* d, uint32_t parent, uint32_t child) {
     }
     p->last_child = child;
     p->count++;
+    /* every link changes SOME container's child list (#377's cache) */
+    d->child_cache_id = UINT32_MAX;
+    d->child_cache_len = 0;
 }
 
 /* ---- mutation side tables (64-2a) ---- */
@@ -485,6 +488,9 @@ yep_dom* yep_dom_create(const yep_allocator* sys) {
      * document that decode-only consumers (CBOR load, serialize, free)
      * never touch; yep_dom_handles() materializes it on first use */
     d->handles = NULL;
+    d->child_cache = NULL;
+    d->child_cache_id = UINT32_MAX;
+    d->child_cache_len = 0;
     if (yep_mutex_init(&d->midx.mu) != 0) {
         yep_pool_destroy(pool);
         yep_free(sys, d);
@@ -513,6 +519,69 @@ struct yep_hpool* yep_dom_handles(yep_dom* d) {
     return h;
 }
 
+/* #377 (ruby #168): seq_at/map_at were O(i) sibling walks — the
+ * bindings' per-element loops made an 80k-row sequence quadratic
+ * (the relaton index: 238-328s where stdlib takes ~3s). The bulk
+ * drain (yeptris_node_children) fixed the bindings; THIS fixes the
+ * accessor: the last indexed container's child ids cache under the
+ * lazy-init mutex, O(1) after the first call. Invalidated by every
+ * child-list mutation (dom_invalidate_child_cache). */
+void dom_invalidate_child_cache(yep_dom* d) {
+    if (d == NULL) {
+        return;
+    }
+    d->child_cache_id = UINT32_MAX;
+    d->child_cache_len = 0;
+}
+
+/* Returns child id at index (UINT32_MAX when out of range); the
+ * container's child count rides *count_out. */
+uint32_t dom_indexed_child(yep_dom* d, uint32_t cid, size_t index, uint32_t* count_out) {
+    const yep_dnode* n = yep_dom_node(d, cid);
+    if (n == NULL) {
+        *count_out = 0;
+        return UINT32_MAX;
+    }
+    *count_out = n->count;
+    if (d->child_cache_id == cid) {
+        if (index < d->child_cache_len) {
+            return d->child_cache[index];
+        }
+        return UINT32_MAX;
+    }
+    /* miss: build once under the lazy-init mutex (Threads.
+     * ReadOnlySharing — concurrent first calls race here exactly like
+     * the handle pool's) */
+    yep_mutex_lock(&d->midx.mu);
+    if (d->child_cache_id != cid) {
+        yep_free(d->sys, d->child_cache);
+        d->child_cache = NULL;
+        d->child_cache_len = 0;
+        if (n->count > 0) {
+            uint32_t* arr = yep_alloc(d->sys, n->count * sizeof(uint32_t));
+            if (arr != NULL) {
+                uint32_t k = 0;
+                for (uint32_t id = n->first_child; id != UINT32_MAX && k < n->count;) {
+                    const yep_dnode* cn = yep_dom_node(d, id);
+                    if (cn == NULL) {
+                        break;
+                    }
+                    arr[k++] = id;
+                    id = cn->next_sibling;
+                }
+                d->child_cache = arr;
+                d->child_cache_len = k;
+            }
+        }
+        d->child_cache_id = cid;
+    }
+    yep_mutex_unlock(&d->midx.mu);
+    if (index < d->child_cache_len) {
+        return d->child_cache[index];
+    }
+    return UINT32_MAX;
+}
+
 void yep_dom_destroy(yep_dom* d) {
     if (d == NULL) {
         return;
@@ -523,6 +592,7 @@ void yep_dom_destroy(yep_dom* d) {
     yep_free(d->sys, d->mut_att);
     yep_free(d->sys, d->mut_depth);
     yep_hpool_destroy(d->handles);
+    yep_free(d->sys, d->child_cache);
     yep_pool_destroy(d->pool);
     yep_free(d->sys, d);
 }

data/vendor/libyeptris/src/yeptris/dom/dom.h

@@ -114,6 +114,13 @@ typedef struct yep_dom {
      * YeptrisNode wrappers here, so read-only document sharing across
      * threads is safe; parse-path pools stay single-threaded */
     struct yep_hpool* handles;
+    /* #377: the lazy indexed-children cache — child ids of the LAST
+     * container indexed via seq_at/map_at. Built once per container
+     * under the lazy-init mutex (read-only sharing), O(1) after;
+     * invalidated by every child-list mutation. NULL when idle. */
+    uint32_t* child_cache;
+    uint32_t child_cache_id; /* UINT32_MAX = empty */
+    uint32_t child_cache_len;
     yep_dnode* nodes;
     uint32_t ncount, ncap;
     uint32_t* docs; /* document root node ids */
@@ -200,6 +207,9 @@ void dom_mut_set_depth(yep_dom* d, uint32_t id, uint16_t depth);
 struct yep_hpool* yep_hpool_create(const yep_allocator* sys);
 /* Lazy handle-pool acquisition (#157): NULL dom or OOM stays NULL. */
 struct yep_hpool* yep_dom_handles(yep_dom* d);
+/* #377: the lazy indexed-children cache (seq_at/map_at's O(1) leg) */
+void dom_invalidate_child_cache(yep_dom* d);
+uint32_t dom_indexed_child(yep_dom* d, uint32_t cid, size_t index, uint32_t* count_out);
 void yep_hpool_destroy(struct yep_hpool* p);
 void* yep_hpool_alloc(struct yep_hpool* p, size_t size, size_t align);
 

data/vendor/libyeptris/src/yeptris/dom/mutate.c

@@ -268,6 +268,8 @@ static int unlink_child(yep_dom* d, uint32_t parent, uint32_t child) {
             p->last_child = c;
         }
     }
+    d->child_cache_id = UINT32_MAX; /* #377: the list changed */
+    d->child_cache_len = 0;
     d->nodes[child].next_sibling = UINT32_MAX;
     dom_mut_set_att(d, child, 0);
     p->count--;

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

@@ -646,11 +646,12 @@ YEPTRIS_API YeptrisNode yeptris_node_seq_at(YeptrisNode handle, size_t index) {
     if (n == NULL || n->kind != YEP_DOM_SEQUENCE || index >= n->count) {
         return NULL;
     }
-    const yep_dom* dom = ((yeptris_node*)handle)->doc->dom;
-    uint32_t id = n->first_child;
-    for (size_t i = 0; i < index && id != UINT32_MAX; i++) {
-        const yep_dnode* cur = yep_dom_node(dom, id);
-        id = cur ? cur->next_sibling : UINT32_MAX;
+    yep_dom* dom = ((yeptris_node*)handle)->doc->dom;
+    uint32_t count = 0;
+    /* the node id IS its dnode index */
+    uint32_t id = dom_indexed_child(dom, ((yeptris_node*)handle)->id, index, &count); /* #377 */
+    if (id == UINT32_MAX) {
+        return NULL;
     }
     return wrap((yeptris_node*)handle, id);
 }
@@ -685,10 +686,11 @@ YEPTRIS_API int yeptris_node_map_at(YeptrisNode handle, size_t index, YeptrisNod
         return -1;
     }
     yeptris_node* h = (yeptris_node*)handle;
-    const yep_dom* d = h->doc->dom;
-    uint32_t child = n->first_child;
-    for (size_t i = 0; i < index * 2; i++) {
-        child = d->nodes[child].next_sibling; /* pairs are key,value,… */
+    yep_dom* d = h->doc->dom;
+    uint32_t count = 0;
+    uint32_t child = dom_indexed_child(d, h->id, (size_t)index * 2, &count); /* #377 */
+    if (child == UINT32_MAX) {
+        return -1;
     }
     if (key != NULL) {
         *key = wrap(h, child);

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: yeptris
 version: !ruby/object:Gem::Version
-  version: 0.6.15.2
+  version: 0.6.16.1
 platform: aarch64-linux
 authors:
 - Ribose Inc.