diff --git a/Benchmarks/bench-libxml2.c b/Benchmarks/bench-libxml2.c index bca4b01..260630a 100644 --- a/Benchmarks/bench-libxml2.c +++ b/Benchmarks/bench-libxml2.c @@ -61,21 +61,31 @@ int main(int argc, char **argv) { } printf("libxml2,serialize,%ld,%.6f\n", size, best_serialize); - /* XPath: count a broad selection. */ + /* XPath: count a broad selection. libxml2 caps node sets at + * XPATH_MAX_NODESET_LENGTH (10M): past that the evaluation returns + * NULL, which is reported rather than dereferenced. */ double best_xpath = 1e9; - long count = 0; + long count = -1; for (int i = 0; i < iterations; i++) { double t0 = now(); xmlXPathContextPtr ctx = xmlXPathNewContext(doc); xmlXPathObjectPtr res = xmlXPathEvalExpression((const xmlChar *)"count(//item[@kind='even'])", ctx); - count = (long)res->floatval; - double dt = now() - t0; - if (dt < best_xpath) best_xpath = dt; - xmlXPathFreeObject(res); + if (res) { + count = (long)res->floatval; + double dt = now() - t0; + if (dt < best_xpath) best_xpath = dt; + xmlXPathFreeObject(res); + } xmlXPathFreeContext(ctx); + if (!res) break; + } + if (count >= 0) { + printf("libxml2,xpath,%ld,%.6f\n", size, best_xpath); + fprintf(stderr, "libxml2 xpath count: %ld\n", count); + } else { + printf("libxml2,xpath,%ld,refused\n", size); + fprintf(stderr, "libxml2 xpath: evaluation refused (node-set cap)\n"); } - printf("libxml2,xpath,%ld,%.6f\n", size, best_xpath); - fprintf(stderr, "libxml2 xpath count: %ld\n", count); xmlFreeDoc(doc); xmlCleanupParser(); diff --git a/Benchmarks/bench-purexml.swift b/Benchmarks/bench-purexml.swift index 9a5e447..5b69026 100644 --- a/Benchmarks/bench-purexml.swift +++ b/Benchmarks/bench-purexml.swift @@ -29,11 +29,19 @@ func best(_ iterations: Int, _ body: () throws -> Void) rethrows -> Double { // Parse. var tree: PureXML.Model.TreeNode! let parseTime = try best(iterations) { - tree = try PureXML.parseTree(source) + tree = try PureXML.parseTree(source, limits: .init(allowDoctype: true)) } print("purexml,parse,\(bytes),\(String(format: "%.6f", parseTime))") +/// SAX (event stream only): isolates scanner cost from tree construction. +let saxTime = try best(iterations) { + var reader = PureXML.events(source, limits: .init(allowDoctype: true)) + while try reader.next() != nil {} +} + +print("purexml,sax,\(bytes),\(String(format: "%.6f", saxTime))") + // Serialize. let node = tree.node var serialized = "" diff --git a/CHANGELOG.md b/CHANGELOG.md index d4400ca..413ed8a 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -11,6 +11,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0 - Scanner-side validity findings are now located structured values (validation-framework audit follow-up): `DocumentType.validityFindings` carries `ValidityFinding` values (reason plus the declared name they are about) instead of strings, `DTDSchema.declarationErrors` holds ready `ValidationError`s, and every declaration-level finding renders at its declaration's coding path (`x`, `r/@a`) instead of "at root of document", matching the XSD consistency-errors precedent. Finding texts are unchanged. +### Added + +- Public streaming surface (#142) and a public-API test target: `EventReader.next()`, both initializers, and the `Event` alias are public, so external consumers can pull events with bounded memory over arbitrarily large inputs (an internal typealias in a split file had been shadowing the public `Parsing.Event`, narrowing `next()` to internal; the benchmark harness found it). A new `PureXMLPublicAPITests` target compiles against the public surface without `@testable`, so any public facade over internal members fails CI; it pins parse/query/serialize, streaming pull, and the chunked pull source. + +### Fixed + +- XPath document-order sorting is benchmarked at scale across generated (4/42/210 MB) and real corpora (NASA ADC 25 MB, SwissProt 115 MB), all linear: the quadratic fix's own first attempt (an eager sibling table built per predicate evaluation) was caught by sampling the live run and replaced with a lazy table (single-node sets never touch order keys; a parent's table builds on its second lookup). XPath at 42 MB drops 27.1 s to 1.57 s (ratio 190x to 11.7x); at 210 MB PureXML completes in 9.4 s the query libxml2 refuses at its 10M node-set cap (the configurable protective budget is #143). The benchmark harness gains a size series, a CORPUS mode for real-world files, a SAX stage, and a refusal-aware comparison table. +- XPath document-order sorting is no longer quadratic on flat fan-outs (#139, found by the large-corpus benchmark): document-order keys looked up each node's index by scanning its parent's children, so sorting a node-set drawn from one wide element (the 42 MB corpus: 200k siblings) cost 27 seconds. A per-operation sibling-index cache enumerates each distinct parent once, making key computation linear in the tree. + ### Changed - Byte-backed Reader, performance passes 2 and 3 (#139): String input now copies its UTF-8 once into owned storage (an unsafe pointer encapsulated in a class lifetime, never escaping the parser) and scalars decode straight off the pointer, eliminating the per-character closure chain; ASCII literals match and consume at the byte level; and plain-ASCII character data, the bulk of any document, scans as a single byte run with one String built per run. Cumulative parse time on the 4.2 MB benchmark corpus drops 32 percent (0.49 s to 0.34 s, libxml2 ratio 30x to 20x). The W3C conformance suite caught one boundary in the run scanner (a run starting with '>' could swallow the "]]>" check after the character path consumed "]]"; ibm14n01); the run scanner now leaves a leading '>' to the character path, with the case pinned as a unit regression. All suites (W3C xmlconf, RELAX NG, xalan), Linux, and WASM verified. diff --git a/Package.swift b/Package.swift index 199109d..39f1562 100644 --- a/Package.swift +++ b/Package.swift @@ -16,6 +16,11 @@ let package = Package( name: "PureXML", path: "Sources", ), + .testTarget( + name: "PureXMLPublicAPITests", + dependencies: ["PureXML"], + path: "PublicAPITests", + ), .testTarget( name: "PureXMLTests", dependencies: ["PureXML"], diff --git a/PublicAPITests/PublicSurfaceTests.swift b/PublicAPITests/PublicSurfaceTests.swift new file mode 100644 index 0000000..398df3e --- /dev/null +++ b/PublicAPITests/PublicSurfaceTests.swift @@ -0,0 +1,51 @@ +// Deliberately NOT @testable: these tests compile against the public +// surface exactly as an external package would, so a public facade over +// internal members (the #142 class of gap: a public EventReader whose +// next() was internal) fails CI instead of hiding behind testability. +import PureXML +import Testing + +@Suite("Public API surface") +struct PublicSurfaceTests { + @Test("Parse, query, and serialize work through the public surface") + func test_parseQuerySerialize() throws { + let document = try PureXML.parse("xy") + let query = try PureXML.XPath.Query("count(//a)") + #expect(try query.number(over: document) == 2) + #expect(PureXML.serialize(document).contains("x")) + } + + @Test("The streaming reader pulls events publicly, bounded memory") + func test_streamingPull() throws { + var reader = PureXML.events("x") + var names: [String] = [] + var texts: [String] = [] + var comments = 0 + while let event = try reader.next() { + switch event { + case let .startElement(name, _): names.append(name.description) + case let .characters(text): texts.append(text) + case .comment: comments += 1 + default: break + } + } + #expect(names == ["r", "a"]) + #expect(texts == ["x"]) + #expect(comments == 1) + } + + @Test("The streaming reader accepts a chunked pull source") + func test_streamingFromChunks() throws { + var chunks = ["", "hello", ""].makeIterator() + var characters: [Character] = [] + var reader = PureXML.Parsing.EventReader(pulling: { + if characters.isEmpty, let chunk = chunks.next() { characters = Array(chunk) } + return characters.isEmpty ? nil : characters.removeFirst() + }) + var events = 0 + while try reader.next() != nil { + events += 1 + } + #expect(events == 5) + } +} diff --git a/README.md b/README.md index 3e16555..47a2b43 100644 --- a/README.md +++ b/README.md @@ -125,7 +125,7 @@ flowchart TB PureXML is a working, dependency-free XML library today: parse, emit, validate, query, and stream documents on macOS, Linux, Windows, and WASM. The test suite -currently runs **981 tests in 167 suites** (`swift test`). +currently runs **984 tests in 168 suites** (`swift test`). ### Shipped (libxml2-aligned surface) diff --git a/Sources/Parsing/EventReader.swift b/Sources/Parsing/EventReader.swift index 382e18c..ceceb1a 100644 --- a/Sources/Parsing/EventReader.swift +++ b/Sources/Parsing/EventReader.swift @@ -40,7 +40,7 @@ public extension PureXML.Parsing { /// Creates a reader over a character-pulling source. The closure returns /// the next character or nil at end of input. - init( + public init( pulling pull: @escaping () -> Character?, limits: Limits = .default, resolver: EntityResolver = .refusing, @@ -54,7 +54,7 @@ public extension PureXML.Parsing { } /// Creates a reader over a string (a convenience over the streaming init). - init(_ string: String, limits: Limits = .default, resolver: EntityResolver = .refusing, recovering: Bool = false) { + public init(_ string: String, limits: Limits = .default, resolver: EntityResolver = .refusing, recovering: Bool = false) { reader = Reader(string) self.limits = limits self.resolver = resolver @@ -67,7 +67,7 @@ public extension PureXML.Parsing { /// diagnostic and reading continues, popping to a matching open element on /// a mismatched end tag and stopping (so the caller closes what is open) on /// truncation. - mutating func next() throws -> Event? { + public mutating func next() throws -> Event? { if !primed { primed = true if reader.peek() == "\u{FEFF}" { diff --git a/Sources/Parsing/EventReaderEndTag.swift b/Sources/Parsing/EventReaderEndTag.swift index caa8981..81c5400 100644 --- a/Sources/Parsing/EventReaderEndTag.swift +++ b/Sources/Parsing/EventReaderEndTag.swift @@ -1,7 +1,9 @@ /// The end-tag scanner, split from the reader body to keep it under the /// length caps. Behavior is part of EventReader proper. extension PureXML.Parsing.EventReader { - typealias Event = PureXML.Parsing.Event + /// Public: this alias shadows the public Parsing.Event for the whole + /// type, so it must not narrow the public next() signature (#142). + public typealias Event = PureXML.Parsing.Event typealias ParseError = PureXML.Parsing.ParseError mutating func scanEndTag() throws -> Event { let mark = reader.mark diff --git a/Sources/XPath/SiblingIndexCache.swift b/Sources/XPath/SiblingIndexCache.swift new file mode 100644 index 0000000..f01b774 --- /dev/null +++ b/Sources/XPath/SiblingIndexCache.swift @@ -0,0 +1,34 @@ +extension PureXML.XPath { + /// A per-operation cache of each node's index among its parent's + /// children. Document-order keys need the index at every ancestor level; + /// without a cache each lookup scans the sibling list, so sorting a + /// node-set drawn from one flat fan-out is quadratic (the 42 MB + /// benchmark corpus spent 27 seconds there). Each distinct parent is + /// enumerated exactly once. + final class SiblingIndexCache { + private var indexes: [ObjectIdentifier: [ObjectIdentifier: Int]] = [:] + private var scannedOnce: Set = [] + + func index(of child: PureXML.Model.TreeNode, in parent: PureXML.Model.TreeNode) -> Int { + let parentIdentity = ObjectIdentifier(parent) + if let table = indexes[parentIdentity] { + return table[ObjectIdentifier(child)] ?? 0 + } + // The table only pays for itself from the second lookup under + // the same parent: a single lookup (one key over a wide parent, + // the per-predicate case that once built a million-entry table + // per evaluation) scans linearly instead. + guard scannedOnce.contains(parentIdentity) else { + scannedOnce.insert(parentIdentity) + return parent.children.firstIndex { $0 === child } ?? 0 + } + var table: [ObjectIdentifier: Int] = [:] + table.reserveCapacity(parent.children.count) + for (index, node) in parent.children.enumerated() { + table[ObjectIdentifier(node)] = index + } + indexes[parentIdentity] = table + return table[ObjectIdentifier(child)] ?? 0 + } + } +} diff --git a/Sources/XPath/XPathNode.swift b/Sources/XPath/XPathNode.swift index dc5605f..3b86fe0 100644 --- a/Sources/XPath/XPathNode.swift +++ b/Sources/XPath/XPathNode.swift @@ -108,16 +108,37 @@ public extension PureXML.XPath { private static let attributeBand = -1 private static func treePath(_ node: PureXML.Model.TreeNode) -> [Int] { + treePath(node, cache: nil) + } + + /// The root path, with an optional per-operation sibling-index cache: + /// without one, finding a node's index among its siblings is linear, + /// which turns a sort over a flat fan-out (one parent, n children) + /// quadratic. The cache enumerates each distinct parent once, making + /// total key computation linear in the tree. + static func treePath(_ node: PureXML.Model.TreeNode, cache: SiblingIndexCache?) -> [Int] { var path: [Int] = [] var current = node while let parent = current.parent { - let index = parent.children.firstIndex { $0 === current } ?? 0 - path.append(index) + path.append(cache?.index(of: current, in: parent) ?? parent.children.firstIndex { $0 === current } ?? 0) current = parent } return path.reversed() } + /// `documentOrder` with the sibling-index cache applied. + func documentOrder(cache: SiblingIndexCache) -> [Int] { + switch self { + case let .tree(node): + return Self.treePath(node, cache: cache) + case let .attribute(owner, attribute): + let index = owner.attributes.firstIndex { $0.name == attribute.name } ?? 0 + return Self.treePath(owner, cache: cache) + [Self.attributeBand, index] + case let .namespace(owner, prefix, _): + return Self.treePath(owner, cache: cache) + [Self.namespaceBand, prefix.hashValue] + } + } + /// Whether `lhs` precedes `rhs` in document order. Each call recomputes /// both order keys; for sorts and minimums over node-sets, prefer /// ``Swift/Sequence/sortedByDocumentOrder()`` and @@ -141,14 +162,20 @@ extension Sequence { /// (decorate-sort-undecorate) rather than recomputing the root path on every /// comparison, which turns an O(n log n x depth) sort into O(n log n). func sortedByDocumentOrder() -> [PureXML.XPath.Node] { - map { (node: $0, key: $0.documentOrder) } + // Zero- and one-node sets are already sorted: the common predicate + // case must not touch order keys at all. + let nodes = Array(self) + if nodes.count <= 1 { return nodes } + let cache = PureXML.XPath.SiblingIndexCache() + return nodes.map { (node: $0, key: $0.documentOrder(cache: cache)) } .sorted { PureXML.XPath.Node.ordered($0.key, before: $1.key) } .map(\.node) } /// The first node in document order, computing each node's order key once. func firstInDocumentOrder() -> PureXML.XPath.Node? { - map { (node: $0, key: $0.documentOrder) } + let cache = PureXML.XPath.SiblingIndexCache() + return map { (node: $0, key: $0.documentOrder(cache: cache)) } .min { PureXML.XPath.Node.ordered($0.key, before: $1.key) }? .node } diff --git a/docs/benchmarks.md b/docs/benchmarks.md index e90297c..565bf25 100644 --- a/docs/benchmarks.md +++ b/docs/benchmarks.md @@ -20,6 +20,33 @@ cross-implementation correctness check. These are the honest starting numbers the #139 optimization passes burn down; the table updates as passes land, never silently. +## Current record (size series + real corpora, best of 3) + +| corpus | parse | serialize | xpath | +|---|---|---|---| +| generated 4.2 MB | 0.34 s (18.7x) | 0.064 s (5.5x) | 0.15 s (11.9x) | +| generated 42 MB | 3.49 s (18.7x) | 0.63 s (5.6x) | 1.57 s (11.7x) | +| generated 210 MB | 19.3 s (19.7x) | 3.19 s (5.6x) | 9.4 s (libxml2: **refused**) | +| NASA ADC 25 MB (real) | 1.73 s (24.5x) | 0.42 s (8.0x) | 0.76 s (19.4x) | +| SwissProt 115 MB (real) | 11.0 s (22.0x) | 1.77 s (5.5x) | 5.0 s (18.7x) | + +Two findings only the large corpora could surface: + +- PureXML's document-order sort was quadratic on flat fan-outs (42 MB: + 27 s in XPath, ratio 190x). Fixed; and the first fix attempt was itself + quadratic in a worse place (an eager sibling-table built per predicate + evaluation: one hour at 42 MB), caught by sampling the live run. The + cache now early-outs single-node sets and builds a parent's table only + on its second lookup. XPath at 42 MB: 27.1 s -> 1.57 s. +- libxml2 refuses XPath evaluation past its XPATH_MAX_NODESET_LENGTH cap + (10M nodes): the 210 MB corpus query returns NULL by design. PureXML + completes it in 9.4 s. The protective counterpart for hostile input is + tracked as #143. + +Scaling: all three PureXML operations are linear in document size (ratios +flat from 4 MB to 210 MB), and there is no 2 GB integer cliff (64-bit +offsets throughout; libxml2's xmlReadMemory takes an int byte count). + ## Optimization log | pass | change | parse | xpath | serialize | diff --git a/scripts/benchmark.sh b/scripts/benchmark.sh index b0ec311..cbc2d87 100644 --- a/scripts/benchmark.sh +++ b/scripts/benchmark.sh @@ -1,21 +1,25 @@ #!/usr/bin/env bash -# PureXML vs libxml2 benchmark (#139). Generates a deterministic corpus, +# PureXML vs libxml2 benchmark (#139). Generates deterministic corpora, # builds both drivers (PureXML in release, the C driver against the SDK's -# libxml2), runs them over the same file, and prints a comparison table. -# Nothing is vendored; the corpus is regenerated on each run. +# libxml2), runs them over the same files, and prints comparison tables. +# Nothing is vendored; corpora are regenerated on each run. # -# Usage: bash scripts/benchmark.sh [items] [iterations] +# Usage: bash scripts/benchmark.sh ["items items ..."] [iterations] +# CORPUS=/path/to/file.xml bash scripts/benchmark.sh "" [iterations] +# +# The default size series spans ~4 MB, ~42 MB, and ~210 MB; pass a single +# item count to pin one size, or CORPUS to measure a real-world document. set -euo pipefail cd "$(dirname "$0")/.." -ITEMS="${1:-20000}" -ITERATIONS="${2:-5}" +SIZES="${1:-20000 200000 1000000}" +ITERATIONS="${2:-3}" WORK="$(mktemp -d)" trap 'rm -rf "$WORK"' EXIT -# Deterministic corpus: nested records with attributes, text, and namespaces. -python3 - "$ITEMS" > "$WORK/corpus.xml" << 'PYEOF' +generate_corpus() { + python3 - "$1" > "$2" << 'CORPUSEOF' import sys items = int(sys.argv[1]) out = sys.stdout @@ -29,9 +33,8 @@ for i in range(items): out.write(f' line one\nline two for {i}\n') out.write(' \n') out.write('\n') -PYEOF -SIZE=$(wc -c < "$WORK/corpus.xml" | tr -d ' ') -echo "corpus: $ITEMS items, $SIZE bytes, best of $ITERATIONS runs" +CORPUSEOF +} echo "==> building PureXML (release)" swift build -c release > /dev/null @@ -44,22 +47,39 @@ echo "==> building libxml2 driver" SDK="$(xcrun --show-sdk-path)" clang -O2 -I "$SDK/usr/include/libxml2" -lxml2 Benchmarks/bench-libxml2.c -o "$WORK/bench-libxml2" -echo "==> running" -"$WORK/bench-purexml" "$WORK/corpus.xml" "$ITERATIONS" > "$WORK/pure.csv" -"$WORK/bench-libxml2" "$WORK/corpus.xml" "$ITERATIONS" > "$WORK/lib.csv" - -python3 - "$WORK/pure.csv" "$WORK/lib.csv" << 'PYEOF' +run_one() { + local file="$1" label="$2" + echo "" + echo "==> $label ($(wc -c < "$file" | tr -d ' ') bytes, best of $ITERATIONS)" + "$WORK/bench-purexml" "$file" "$ITERATIONS" > "$WORK/pure.csv" + "$WORK/bench-libxml2" "$file" "$ITERATIONS" > "$WORK/lib.csv" + python3 - "$WORK/pure.csv" "$WORK/lib.csv" << 'TABLEEOF' import sys def load(path): rows = {} for line in open(path): library, operation, size, seconds = line.strip().split(",") - rows[operation] = float(seconds) + rows[operation] = None if seconds == "refused" else float(seconds) return rows pure = load(sys.argv[1]) lib = load(sys.argv[2]) print(f"{'operation':<12} {'PureXML':>12} {'libxml2':>12} {'ratio':>8}") for operation in ["parse", "serialize", "xpath"]: - p, l = pure[operation], lib[operation] - print(f"{operation:<12} {p:>11.4f}s {l:>11.4f}s {p / l:>7.2f}x") -PYEOF + p, l = pure.get(operation), lib.get(operation) + if p is None or l is None: + left = f"{p:.4f}s" if p is not None else "refused" + right = f"{l:.4f}s" if l is not None else "refused" + print(f"{operation:<12} {left:>12} {right:>12} {'-':>8}") + else: + print(f"{operation:<12} {p:>11.4f}s {l:>11.4f}s {p / l:>7.2f}x") +TABLEEOF +} + +if [ -n "${CORPUS:-}" ]; then + run_one "$CORPUS" "real corpus: $CORPUS" +else + for items in $SIZES; do + generate_corpus "$items" "$WORK/corpus-$items.xml" + run_one "$WORK/corpus-$items.xml" "generated, $items items" + done +fi