leptris 1.9.107.0 → 1.9.107.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 64e0e821d66fb8c8d25debf605648a6d533f432e0882f1a7cecf58b2e2a19e52
-  data.tar.gz: 70bb6cf535b69d0ae0c470459f183d6420e87db4a8bef12c89152f1d739c793b
+  metadata.gz: 859b03f81870e25ede0d7b9f05ac20d4013506220a69614dc7ea82d9db27b371
+  data.tar.gz: e3eb2b0658fb4698717e61a5a8a70b3195e0f229137b3035abf59f32096c0ab0
 SHA512:
-  metadata.gz: eadc86a700848a3a0980d0e2af292ded87619a4bd958a0ea897a310c35caf00dfca7d4c1a997e1760d48d18872cb84458440601bc437a120131e6f4344a614b0
-  data.tar.gz: 9d6a4fbc7dc3b88e96c896a4b55d10583b9764979a4d7f8eb3d0a51be3422d870f758a805c566529e85c94c2bcd7512b5715479e4d1e04295ce1936ff78599ab
+  metadata.gz: 179b50450680923f7bb71c1079e3226559d04a754ce0892765fd2bcbdbb4d3b7506d88dda7a8f58bd2f4148c6266248d0fe84ff5efd63485b14a047e6835aa1b
+  data.tar.gz: 5a4653e79904ec197c7775e08cd11478d2dc2638a2d28a52b4b901f6cb93745793cd4748bcd397cf92b051349865d5b16741f52e2cb86866c0d3ae129cf4af50

data/CHANGELOG.md

@@ -5,6 +5,25 @@ 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).
 
+## [1.9.107.1] - 2026-09-08
+
+### Improved
+
+- **SAX drain: bulk kind strip** (TODO.restructure/22 — the moxml
+  0.77-0.93x row): the recorder's dispatch loop read the event
+  kind with a `get_uint8` per EVENT (a memory crossing for every
+  record, dispatched or not); the strip now reads in ONE bulk
+  `get_bytes` + ONE `unpack` per drain via the memoized
+  per-count template — kind codes 0..10 unpack to immediate
+  Fixnums, so the change strictly removes ~count FFI crossings
+  with zero new allocations. Allocation profile (load-independent
+  measurement) is far ahead either way: text-only 32,381 vs
+  Nokogiri's 400,061 allocs/parse (12x fewer); all-events 112,440
+  vs 400,034 (3.6x fewer) — the remaining gap is pure per-event
+  CPU, which this addresses. Timing verification gated on a clean
+  host window (bench staged; the shared machine sat at load
+  40-400 today).
+
 ## [1.9.107.0] - 2026-09-08
 
 ### Changed

data/TODO.restructure/22-sax-drain-bulk-strip.md

@@ -0,0 +1,28 @@
+# 22 — SAX drain: bulk kind strip (the 0.77-0.93x row)
+
+Status: CODE DONE — timing verification gated on a clean window
+
+The moxml table shows SAX at 0.77-0.93x Nokogiri (it fully won at
+the 1.9.40-era battery). Investigation under load-404 conditions
+(timing unusable; allocation profiling is load-independent):
+
+- **Allocations are NOT the gap** — measured GC-stat deltas:
+  text-only 32,381 vs Nokogiri's 400,061 allocs/parse (12x fewer);
+  all-events 112,440 vs 400,034 (3.6x fewer).
+- **The time cost was per-event FFI crossings in the drain**: the
+  kind strip was read with a `get_uint8` per EVENT (all events,
+  dispatched or not — a memory crossing each) after the
+  whole-record-unpack was retired for its 2.25M-Integer profile.
+- **Fix**: the kind strip reads in ONE `get_bytes` + ONE `unpack`
+  per drain through the already-defined memoized
+  `KIND_TEMPLATE_FOR` — kind codes 0..10 unpack to IMMEDIATE
+  Fixnums (zero allocation), so this strictly removes ~count FFI
+  crossings per drain with no new allocations. Per-dispatched-event
+  field reads (2-3 `get_uint32`) remain.
+
+- [x] Bulk strip restored; SAX suite green; allocation profile
+      re-verified (32k/112k vs 400k/400k).
+- [ ] **Timing verification** (the actual 0.77-0.93x row): bench
+      staged at /tmp/bench_sax.rb — requires load < 20 (the host
+      has been at 40-400). Gate: text-only and all-events SAX at
+      parity or better vs Nokogiri on the raw binding.

data/lib/leptris/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Leptris
-  VERSION = "1.9.107.0"
+  VERSION = "1.9.107.1"
 end

data/lib/leptris/xml/sax/recorder.rb

@@ -159,16 +159,20 @@ class Leptris::XML::SAX::Recorder
         @handle, len_ptr)
       arena = arena_ptr.read_bytes(len_ptr.read_uint64)
       count = count_ptr.read_uint64
-      # Kind is the first byte of each record — read it off the C
-      # pointer. The previous path copied every record into Ruby
-      # and unpacked a Fixnum array just for the kind strip
-      # (2.25M Integers on a single-feed drain); one get_uint8 per
-      # event replaces both.
+      # Kind strip in ONE bulk read + ONE unpack per drain: the
+      # per-count template reads the first byte of each record,
+      # skipping the stride (KIND_TEMPLATE_FOR is memoized per
+      # count). Kind codes are 0..10 — unpack yields immediate
+      # Fixnums, so the strip allocates nothing. This replaces a
+      # get_uint8 crossing per EVENT (all events, dispatched or
+      # not) with one memory read per drain — the SAX gap to
+      # Nokogiri's per-event C callback lives exactly here.
+      kinds = records_ptr.get_bytes(0, count * RECORD_STRIDE)
+                         .unpack(KIND_TEMPLATE_FOR[count])
       one_arg_start = dispatched[:start_element] == :one_arg
       i = 0
       while i < count
-        code = records_ptr.get_uint8(i * RECORD_STRIDE)
-        kind = KIND_BY_CODE[code]
+        kind = kinds[i] && KIND_BY_CODE[kinds[i]]
         if kind && dispatched[kind]
           base = i * RECORD_STRIDE
           case kind

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: leptris
 version: !ruby/object:Gem::Version
-  version: 1.9.107.0
+  version: 1.9.107.1
 platform: ruby
 authors:
 - Ribose Inc.
@@ -90,6 +90,7 @@ files:
 - TODO.restructure/19-gitignore-hygiene.md
 - TODO.restructure/20-utf8proc-enablement.md
 - TODO.restructure/21-typeddata-variant-prototype.md
+- TODO.restructure/22-sax-drain-bulk-strip.md
 - docs/adr/0001-lockstep-mirror.md
 - docs/adr/0002-utf8-at-the-seam.md
 - docs/adr/0003-readonly-memoization-pattern.md