rusteerio 1.1.0 → 1.1.2

package/Cargo.lock

@@ -720,7 +720,7 @@ dependencies = [
 
 [[package]]
 name = "rusteerio"
-version = "1.1.0"
+version = "1.1.2"
 dependencies = [
  "ego-tree",
  "html5ever",

package/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "rusteerio"
-version = "1.1.0"
+version = "1.1.2"
 edition = "2021"
 description = "Rust HTML5 parser and CSS selector engine for the rusteerio Node.js package"
 license = "MIT"

package/README.md

@@ -177,19 +177,19 @@ identical on all 5,000 pages, and times include loading.
 
 | Measure | Result |
 | --- | --- |
-| Total time | Cheerio 17.8 s, rusteerio 9.1 s: **1.97× faster** |
-| Pages where rusteerio was faster | **4,973 of 5,000** |
-| Per-page speedup, 10th / 50th / 90th percentile | 1.21× / 1.51× / 1.92× |
-| Slowest page | 0.78×, a 4 KB page |
+| Total time | Cheerio 15.9 s, rusteerio 7.7 s: **2.06× faster** |
+| Pages where rusteerio was faster | **4,995 of 5,000** |
+| Per-page speedup, 10th / 50th / 90th percentile | 1.30× / 1.62× / 1.98× |
+| Slowest page | 0.90× |
 
-The 27 slower pages are all small, where a few microseconds decide the ratio.
+The five slower pages (0.90× to 1.00×) measured 1.14× to 2.01× faster when
+rerun with 25 samples each on another machine.
 
 **Single operations.** `bench/operations.mjs` times 25 operations on freshly
 loaded generated documents of 200 and 2,000 cards. Including each engine's load
-time, rusteerio was faster for **48 to 49 of 50** HTML measurements across three
-runs (median 1.81–1.89×). XML results varied more between runs: **29 to 41 of 50**
-(median 1.08–1.38×). HTML loading alone was 1.76–1.96× faster; XML loading of the
-large document ranged from parity to 1.39× faster.
+time, rusteerio was faster for **49 to 50 of 50** HTML measurements across two
+runs (median 1.89–1.90×) and **48 of 50** XML measurements (median 1.84×). HTML
+loading alone was 1.75–1.89× faster and XML loading 1.67–2.26× faster.
 
 **Matched CPU budgets.** `bench/fair.mjs` gives both engines the same enforced
 CPU budget; batch comparisons give Cheerio a reused Node worker pool and include
@@ -198,41 +198,41 @@ Rusteerio was faster in **all 45 cases**. Representative results:
 
 | Workload | CPUs | Cheerio | rusteerio | rusteerio |
 | --- | ---: | ---: | ---: | --- |
-| Load/query, 500 articles | 1 | 38.01 ms | 3.61 ms | 10.52× faster |
-| Load/query, 5,000 articles | 1 | 172.87 ms | 39.17 ms | 4.41× faster |
-| HTML mutations, 5,000 articles | 1 | 249.72 ms | 84.42 ms | 2.96× faster |
-| XML load/query, 5,000 articles | 1 | 69.34 ms | 18.20 ms | 3.81× faster |
-| Extract 16 documents × 5,000 articles | 1 | 3,441.82 ms | 533.58 ms | 6.45× faster |
-| Extract 16 documents × 5,000 articles | 8 | 1,126.61 ms | 179.02 ms | 6.29× faster |
-| Nested callback extraction, 8 × 500 articles | 8 | 78.27 ms | 52.03 ms | 1.50× faster |
-| Cold extraction, 16 × 20 articles | 8 | 179.01 ms | 33.78 ms | 5.30× faster |
-| 8 concurrent batches × 8 × 300 articles | 8 | 125.86 ms | 22.99 ms | 5.47× faster |
-| HTML stream, 50,000 elements | 1 | 1,364.09 ms | 328.79 ms | 4.15× faster |
-| XML stream, 50,000 elements | 8 | 123.17 ms | 100.52 ms | 1.23× faster |
-| Append/update/remove, 4,000 cycles | 1 | 666.87 ms | 458.55 ms | 1.45× faster |
+| Load/query, 500 articles | 1 | 36.53 ms | 3.37 ms | 10.82× faster |
+| Load/query, 5,000 articles | 1 | 152.78 ms | 34.90 ms | 4.38× faster |
+| HTML mutations, 5,000 articles | 1 | 241.28 ms | 68.12 ms | 3.54× faster |
+| XML load/query, 5,000 articles | 1 | 65.28 ms | 15.24 ms | 4.28× faster |
+| Extract 16 documents × 5,000 articles | 1 | 2,874.89 ms | 447.12 ms | 6.43× faster |
+| Extract 16 documents × 5,000 articles | 8 | 805.16 ms | 122.93 ms | 6.55× faster |
+| Nested callback extraction, 8 × 500 articles | 8 | 54.17 ms | 33.09 ms | 1.64× faster |
+| Cold extraction, 16 × 20 articles | 8 | 15.19 ms | 3.17 ms | 4.79× faster |
+| 8 concurrent batches × 8 × 300 articles | 8 | 127.82 ms | 23.94 ms | 5.34× faster |
+| HTML stream, 50,000 elements | 1 | 1,086.17 ms | 297.04 ms | 3.66× faster |
+| XML stream, 50,000 elements | 8 | 122.06 ms | 67.44 ms | 1.81× faster |
+| Append/update/remove, 4,000 cycles | 1 | 613.95 ms | 388.99 ms | 1.58× faster |
 
 Cold measurements include engine import, pool initialization and the first
 batch.
 
 **Where Cheerio can be faster.** Rusteerio builds node objects lazily from the
-native tree, so code that visits every node through DOM properties (walking
-`children` and reading `attribs` of every element) pays for creating those
-objects. Including load, that walk was 0.83–0.99× in HTML and 0.37–0.83× in XML,
-where Cheerio's htmlparser2 parser leaves less load time to win back. On the
-large XML document, wrapping thousands of elements one by one (`$(element)` with
-`.attr()`, `.parents()`, `.siblings()` or `.next()`), appending to thousands of
-elements and rewriting attributes element by element measured 0.58–0.97×. Very
-small pages can measure either way.
+native tree, and an element's `attribs` is a `Proxy` so that assigning or
+deleting a key changes the document, as it does in Cheerio. Code that visits
+every node and enumerates every element's attributes (`Object.keys(node.attribs)`)
+pays for creating those objects and for V8's slower key enumeration on proxies.
+Including load, that walk measured 0.58–0.69× in XML, where Cheerio's
+htmlparser2 parser leaves less load time to win back, and 1.00× or better in
+HTML. Walking the same nodes without enumerating attributes is faster in both.
+Very small pages can measure either way.
 
 **Memory.** A batch extracting 16 documents of 5,000 articles with eight CPUs
-ended at 595 MB RSS for rusteerio and 2,535 MB for Cheerio's worker pool, and the
-HTML stream of 50,000 elements at 491 MB versus 664 MB. The XML stream ended
-higher, at 421 MB versus 218 MB, and the append/update/remove workload about even
-(93 MB versus 90 MB). Linux builds use mimalloc, which keeps freed pages briefly
-for reuse; setting `MIMALLOC_PURGE_DELAY=0` returns them at once and lowered the
-streams to 363 MB (HTML) and 286 MB (XML), at a cost in speed for later
-documents. RSS includes initialization and resident input, and sampled peaks do
-not establish a universal memory bound.
+ended at 581 MB RSS for rusteerio and 2,497 MB for Cheerio's worker pool, the
+HTML stream of 50,000 elements at 491 MB versus 666 MB, and the large XML
+load/query at 140 MB versus 147 MB. The XML stream ended higher, at 285 MB versus
+221 MB, and the append/update/remove workload about even (93 MB versus 91 MB).
+Linux builds use mimalloc, which keeps freed pages briefly for reuse; setting
+`MIMALLOC_PURGE_DELAY=0` returns them at once and lowers stream RSS, at a cost in
+speed for later documents. RSS includes initialization and resident input, and
+sampled peaks do not establish a universal memory bound.
 
 These results describe this VM and revision; hardware, runtime and input shape
 affect them. See [the results](bench/results/2026-10-14/) and

package/THIRD_PARTY_NOTICES.md

@@ -4,7 +4,7 @@ This file preserves third-party license and attribution notices for the Rust nat
 
 Generated by `node scripts/generate-notices.mjs` from `cargo metadata --locked`, crate license files, and the pinned supplemental sources recorded in `scripts/notices/sources.json`. It includes build dependencies and dependencies for all target platforms; a particular binary may use a subset. Identical notice texts are reproduced once and referenced below. The html5ever, markup5ever and html5gum crates contain local observer-hook and opt-in UTF-16 transport patches; their complete patched source and original notices are included in `native/vendor`. Other dependency source files are unmodified.
 
-Cargo.lock SHA-256: `a7962957958e0640ea9e40f025e877864c03bd5934a1ca2e19df80a5481d72ff`.
+Cargo.lock SHA-256: `b834f3bb9a4bb9e7889f20f789f80af645271bcfd0057cf4217843b17e3f5ae5`.
 
 ## Source availability