leptris 1.9.40-aarch64-linux → 1.9.42-aarch64-linux

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 1753f68ffbc65299a2742ad1645a74e8feb3c054cf25a44ab01b8571aaa78e1f
-  data.tar.gz: 9a4076ab46da988a84261485317b153f33220878851bc95e586aaebc9df2a524
+  metadata.gz: 7b71e69e85187739869f3d897256675673d450e21534061329aa6f1ccb86005d
+  data.tar.gz: 2511132c3fb1bb8640b4419f528f83769aaced64e37ed3f7ff7a4892716476de
 SHA512:
-  metadata.gz: 3ff2d04232695b730e659859a68d974c9d4f52fc87e7b41b0ba2290d8e0e99a04d62933cacd060b53c086f38497ed2224680e5cd224751fdef48d3ef9f90d0b6
-  data.tar.gz: 53112f18a2ba2ce418433d4bdb18bb435f60a701be8d02d44736e0da9ef956aa2965c594a533bec5f72bbd09fef4dc393722afa4454067c5b97829716ae01066
+  metadata.gz: a74c1bd76a6697f6e3a65084b7a94120f7dff07460065e80c26540b03e06133df3b78d399325a939a1b80bcfdd4cb3bf4b4b26ff642ee45586bca2bdd88a79bb
+  data.tar.gz: 23d7e78e096d3c68b5e3c7c726faaf14d817f8b3f5eb81c56ee3094d962e463d719cb54847b38e24bdf8c4155b030ca29beeb2c48abafed3b894abc200c83bc0

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: aarch64-linux
 authors:
 - Ribose Inc.