leptris 1.9.40 → 1.9.42

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 63ca61bcacf68aee02e3fb94c53f84d64af4474cf5b250482135a632b5ac2009
-  data.tar.gz: 269dd93a547a099a5eb0685f9d58ecfa2d281d373f969cc8e69a7cd6aece5614
+  metadata.gz: 5de687a1b7597e0bdf6ff80f85566435d9927e136a86ec758a554928d85910d2
+  data.tar.gz: 9171a19d4620e4d482c573e1f21e76a8ee253b175a8df34569e9711f1fcc42e4
 SHA512:
-  metadata.gz: d908e6a85d409bae57e50703a87be059e984a67af44139769475f6baa82631894d9cbd26390442f32af54d2987242fee8bc528f70f52e0284692fd522cb2ea23
-  data.tar.gz: 9826c4922df24a0703eca6e39b8b6f62fe5c6deccdf5212b7e8bcc76897b74ce4cc587add6a95aa60324dcff538087cdc5bb5ba35e421a37ce0bc26a18237bad
+  metadata.gz: b2dadeb0a139c9a6396dd82defd080dc014e056a8959321a7f7dff13851503d8d53b6a7b1dc4f987a4bb131ce9054e3c0073d7234d6a8ff0e73caae2fc71cf7e
+  data.tar.gz: 1648229492701f2805e405475b7abf4f59108844d79da56118f47e15fdb048ebf03f5b7a1d4c3f48c566e2c13b1881b50b7a09df9512fcb2cd928ae045260f3a

data/CHANGELOG.md

@@ -5,6 +5,37 @@ All notable changes to Leptris will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.9.42] - 2026-08-30
+
+### Fixed
+
+- **Pull batch staging corruption: fail loudly, never deliver
+  garbage (leptris/leptris#646)**: `leptris_pull_next_batch`'s
+  staging arena misplaces record strings once staged content
+  crosses its block boundary (~185 bytes of attribute value under
+  nested attr-carrying ancestors — the #625 fix family did not
+  reach the batch staging). `Parser#each_batch` now validates each
+  staged record — unknown type codes, empty/control-byte/invalid-
+  encoding names cannot be legal XML events — and raises a
+  descriptive error pointing at `#each` (the cursor path reads the
+  same documents correctly; SAX default, recorder drain, and DOM
+  are unaffected). Bisection attached upstream: ok at 180 staged
+  bytes, corrupt at 190, in the minimized shape.
+
+## [1.9.41] - 2026-08-30
+
+### Changed
+
+- **One thread-local access per children batch**: fetch_children
+  re-looked up the scratch hash inside the ints helper on every
+  call; both buffers now resolve through a single access (and the
+  kinds buffer sizes from the pointers capacity directly). Cold
+  full-tree walk: 254 -> 247 ms per big.xml walk (the gap vs
+  Nokogiri narrows 1.66x -> 1.56x — the remainder is the binding's
+  per-node allocation floor, re-measured and filed upstream as
+  leptris/leptris#645 together with the unchanged 1.64x scalar
+  XPath gap from #617's other half).
+
 ## [1.9.40] - 2026-08-30
 
 ### Changed

data/lib/leptris/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Leptris
-  VERSION = "1.9.40"
+  VERSION = "1.9.42"
 end

data/lib/leptris/xml/ffi.rb

@@ -911,17 +911,22 @@ module Leptris
       # (leptris_node_children_ex, 1.9.18 #617) so wrappers skip
       # the per-child get_type dispatch.
       def self.fetch_children(c_ptr)
+        # One thread-local access serves both buffers (the helper
+        # re-looked it up); the kinds buffer sizes from the POINTERS
+        # buffer's capacity — it may have grown past FETCH_INITIAL on
+        # earlier use, and C writes one kind per copied child.
         scratch = (Thread.current[:leptris_scratch] ||= {})
         buffer = scratch[:pointers]
         if buffer.nil?
           buffer = scratch[:pointers] =
             ::FFI::MemoryPointer.new(:pointer, FETCH_INITIAL)
         end
-        # size the kinds buffer from the POINTERS buffer's capacity —
-        # it may have grown past FETCH_INITIAL on earlier use, and C
-        # writes one kind per copied child
         cap = buffer.size / PTR_BYTES
-        kinds = scratch_ints(cap)
+        kinds = scratch[:ints]
+        if kinds.nil? || kinds.size / INT_BYTES < cap
+          kinds&.free
+          kinds = scratch[:ints] = ::FFI::MemoryPointer.new(:int, cap)
+        end
         copied = leptris_node_children_ex(c_ptr, buffer, kinds, cap)
         if copied == cap
           total = leptris_node_children_ex(c_ptr, nil, nil, 0)
@@ -929,7 +934,9 @@ module Leptris
             buffer.free
             buffer = scratch[:pointers] =
               ::FFI::MemoryPointer.new(:pointer, total)
-            kinds = scratch_ints(total)
+            kinds.free
+            kinds = scratch[:ints] =
+              ::FFI::MemoryPointer.new(:int, total)
             copied = leptris_node_children_ex(c_ptr, buffer, kinds, total)
           end
         end

data/lib/leptris/xml/pull.rb

@@ -85,14 +85,28 @@ module Leptris::XML::Pull
           name_ptr = buffer.get_pointer(base + name_off)
           text_ptr = buffer.get_pointer(base + text_off)
           type = TYPES[buffer.get_int(base + type_off)]
+          name = name_ptr.null? ? nil :
+            name_ptr.read_string.force_encoding(Encoding::UTF_8)
+          # Staging-corruption guard (leptris/leptris#646): the C
+          # batch's staging arena misplaces record strings once the
+          # staged content crosses its block boundary (~185 bytes of
+          # attribute value under nested attr-carrying ancestors) —
+          # names come back empty, control-byte-laced, or
+          # invalid-encoding, and unknown type codes surface as nil.
+          # Element names can never look like that: fail loudly
+          # instead of delivering garbage, and point at the cursor
+          # path, which reads the same documents correctly.
+          if type.nil? ||
+             (name && (name.empty? || !name.valid_encoding? ||
+                       name.match?(/[\x00-\x08\x0e-\x1f]/)))
+            raise Leptris::XML::Error,
+              "pull batch staging corruption (leptris/leptris#646) — " \
+              "use Parser#each (cursor) for this document"
+          end
           text = text_ptr.null? ? nil :
             text_ptr.read_string.force_encoding(Encoding::UTF_8)
           text = Leptris::XML::FFI.read_pi_data(text) if type == :pi
-          Event.new(type,
-            name_ptr.null? ? nil :
-              name_ptr.read_string.force_encoding(Encoding::UTF_8),
-            text, nil
-          )
+          Event.new(type, name, text, nil)
         end
         # The attr mirror holds the batch's most recent start.
         last_start = nil

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: leptris
 version: !ruby/object:Gem::Version
-  version: 1.9.40
+  version: 1.9.42
 platform: ruby
 authors:
 - Ribose Inc.