leptris 1.9.41 → 1.9.43

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: a670640aa8b76b1d6f14f1e50b54ac329a477a6f293dc17e710524822690b0d8
-  data.tar.gz: cfc26d52da5bba121bfbd2d3424b4aa64acd6feb1c2a9be143f99002d16aa1fc
+  metadata.gz: a87e2a5aa4eddbbaf1140aa87d33fcd3e0dccfb7ea505cbcaf917dbd3bf715fd
+  data.tar.gz: '09b6e2b0d46ed3235945f3ad3355fbf46d515ddfc0b3bce2ac75468a80d0af5d'
 SHA512:
-  metadata.gz: c2707588ab5775016e0bafd1554f05e872ae98f35cd9bf347471172f1f290f0daa2badda34c0dbdefdcd4c959e5ff3b5855ff1c20c034e91cbf3d508559fda74
-  data.tar.gz: 5ecbe0965f5e489a9e9eae2fab172a6826a0680ca64856d1a025d822b62c1e88c2badcacd5d6540b5e59144ae6ce5c0311d3e67d012afda5cc9ec0f7804fe566
+  metadata.gz: 267bbe8cac5597fa0e40736f087fffc739c9f3c1fb3063fbe652b25bb9e0a958424b283a5933a963c919d20db6128f35dcd2c98b94439613eeed3edcd17877f0
+  data.tar.gz: 93075be6a11e7fe9cdebfd812c617fe274ef9f93d67cce33d0c83b6e173b9f1fe6515aaff73907155fd09bae9d4867b8e6e35710551764b5554ea8c1214eefe7

data/CHANGELOG.md

@@ -5,6 +5,49 @@ 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).
 
+## [1.9.43] - 2026-08-31
+
+### Fixed
+
+- **Pull prefix events surface (leptris-ruby#648's residual)**: the
+  engine's pull stream has always carried start/end prefix events
+  (codes 8/9; the default namespace's prefix is legitimately `""`)
+  — the binding's type map lacked the kinds, silently delivering
+  nil-typed Events that the 1.9.42 staging guard then misread as
+  corruption on healthy engines. `:start_prefix`/`:end_prefix` now
+  map correctly on both cursor and batch paths, and the guard's
+  name check only invalidates kinds that must carry names
+  (start/end element, PI target). Engine 1.9.21's staging fix
+  (upstream #648) verified clean through the batch path once the
+  kinds were mapped.
+
+### Meta
+
+- The 1.9.19–1.9.21 lockstep is HELD: engine 1.9.19+ carries a DOM
+  parse regression — a self-closing element followed by text inside
+  a depth-2 parent fails to parse (`<div><p><br/>hello</p></div>`)
+  — filed as leptris/leptris#653. `Node#visit` (the #645a
+  wrap-free walk), the sized ext-serialize entry (PR #107's
+  substance), and this pull fix's engine-1.9.21 verification are
+  parked on a branch awaiting the fix.
+
+## [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

data/lib/leptris/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Leptris
-  VERSION = "1.9.41"
+  VERSION = "1.9.43"
 end

data/lib/leptris/xml/ffi.rb

@@ -766,6 +766,10 @@ module Leptris
       PULL_PI            = 5
       PULL_END_DOCUMENT  = 6
       PULL_ERROR         = 7
+      # Prefix events (types.h LeptrisPullEventType 8/9): the
+      # default namespace's prefix is legitimately "".
+      PULL_START_PREFIX  = 8
+      PULL_END_PREFIX    = 9
 
       LEPTRIS_PARSE_DEFAULT = 0
       LEPTRIS_PARSE_DROP_WS_TEXT = 1

data/lib/leptris/xml/pull.rb

@@ -28,6 +28,13 @@ module Leptris::XML::Pull
     Leptris::XML::FFI::PULL_PI => :pi,
     Leptris::XML::FFI::PULL_END_DOCUMENT => :end_document,
     Leptris::XML::FFI::PULL_ERROR => :error,
+    # Prefix events ride the stream ("" prefix = the default
+    # namespace): the engine has always delivered them; the map
+    # lacked the kinds, silently yielding nil-typed Events — which
+    # the staging guard then misread as corruption after the real
+    # engine fix landed (leptris/leptris#648's residual).
+    Leptris::XML::FFI::PULL_START_PREFIX => :start_prefix,
+    Leptris::XML::FFI::PULL_END_PREFIX => :end_prefix,
   }.freeze
 
   class Parser
@@ -85,14 +92,30 @@ 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#648): 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.valid_encoding? ||
+                       name.match?(/[\x00-\x08\x0e-\x1f]/))) ||
+             (%i[start_element end_element pi].include?(type) &&
+              (name.nil? || name.empty?))
+            raise Leptris::XML::Error,
+              "pull batch staging corruption (leptris/leptris#648) — " \
+              "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: 1.9.41
+  version: 1.9.43
 platform: ruby
 authors:
 - Ribose Inc.