The attribute-value scanner ran the entity decoder on every value, and the
decoder re-scanned the whole value for '&' before doing nothing when none
was present. The byte-run scanner already visits every value byte, so it
now notes any '&' and an ampersand-free value (the common case) returns its
normalized run directly. Measured against a same-machine-state baseline,
every run came in below every baseline run, ~4% faster parse on the
20k-item generated corpus.
Move the bulk byte-run scanners (contentRunBytes, attributeRunBytes) into
ReaderByteRuns.swift and centralize their cursor mutation through a single
advanceByteRun method, so Reader's position fields stay write-encapsulated
in Reader.swift while both files stay under the length cap. The byte and
Character paths are held to identical decoded values by the differential
oracle, extended with ampersand-free, entity, and character-reference
attribute cases. No spec behavior changed.
Keep the reader on its byte fast path through dispatch and name/attribute
scanning so plain-ASCII documents avoid the per-Character String machinery
(String.Iterator.next, string comparison, small-string materialization,
bridge-object ARC) that a profile indicted as the parse bottleneck.
The root cause was that a single buffered lookahead Character defeated every
existing byte path, each of which requires an empty buffer, and the content,
markup, and attribute dispatch each peeked a Character before scanning. Add a
non-buffering peekByte for dispatch, a byte-mode skipSpace, an all-ASCII
takeASCIIName, and an attributeRunBytes bulk scanner (with the 3.3.3
whitespace-to-space transform, deferring any carriage return to the Character
path so CRLF is not double-counted). Detect a text run's closing '<' via
peekByte instead of buffering it, which had cascaded the next element's whole
dispatch and name scan onto the slow path. Skip reference decoding for
ampersand-free text. The Character path remains the exact fallback for a
streaming source, spliced entity text, and any non-ASCII byte; the two paths
share scanContentMarkup so they cannot diverge.
Measured parse 20.1x -> 9.5x slower than libxml2 on the 20k-item generated
corpus, roughly a 53 percent wall-clock reduction. A differential oracle pins
the byte and Character paths to identical event streams and error marks over
an adversarial corpus, alongside boundary tests. Verified against the full
local suite, the Apache Xalan XSLT gold-output corpus, the W3C XML conformance
corpora, the RELAX NG spec suite, and the WASM build. No spec behavior changed.
The eduni sections run with exact classified baselines. Fixes: 3.3.3
CDATA attribute-value normalization (literal whitespace to space before
reference decoding; char-ref whitespace survives; tokenized collapse on
SPACE only); production-68 lenient undeclared entities in content and
attribute values (validity finding, reference kept literal, nothing
fetched); errata E15 (references/comments/PIs/whitespace in EMPTY,
char-ref whitespace as data in element content, double-escape shape
included); errata E2 duplicate enumeration tokens; errata E14
declarations completed inside PE replacements (VC: Proper
Declaration/PE Nesting); NCName ID values; namespace declarations
subject to the every-attribute-declared rule. Reader validity machinery
split to EventReaderValidity for caps.
Closes#127
The scanner records declaration provenance (internal-subset entities,
element models, attribute lists); the validator reports the three
standalone='yes' dependencies on the external subset: a default that
would be supplied, a non-CDATA normalization that would change a value,
and whitespace in externally-declared element content. The parser hides
externally-declared entities from standalone documents, so referencing
one reports as undeclared (WFC: Entity Declared, 2.9). The Sun section
is fully clean: not-wf 56/56, valid 28/28, invalid 74/74, no baseline.
Closes#134Closes#124
Make the DTD declaration grammar strict: <!ENTITY> requires its whitespace
(after the keyword, around %, after the name), a real name (not #DEFAULT),
nothing but '>' after the value, NDATA preceded by whitespace and forbidden
on parameter entities; entity-value literals are validated (XML Chars only,
every & a complete reference); SYSTEM/PUBLIC external identifiers parse
strictly (PUBLIC needs both literals for entities and the DOCTYPE, public-only
stays legal for notations) with PubidChar validation; <!ATTLIST> bodies are
validated against XML 1.0 productions 52-60 via a new DTDAttListGrammar
(spacing, attribute types, enumerations, quoted defaults); the reserved xml
PI target is rejected inside the subset. Also: strict UTF-8 decoding (no
silent replacement of malformed sequences), XML Char checks in attribute
values and CDATA, the declaration's version/encoding value grammar, and
required whitespace between pseudo-attributes. The strict machinery lives in
DoctypeScannerStrict/DTDAttListGrammar to stay under the length caps.
W3C xmltest: not-wf/sa 162/186 rejected (from 129; 80 at first contact),
valid/sa 118/119, zero false positives across all 846 tests at every step.
The remaining 24-case baseline is exactly: the entity replacement-text class
(19) and the deliberate internal-subset extensions this package supports as
features (conditional sections, PEs in the internal subset: 5).
Part of #120, #105
Add five well-formedness checks the W3C xmltest suite surfaced: '--' inside
comments (also rejecting the ---> ending), raw '<' in attribute values, the
literal ]]> sequence in character data, attributes not separated by
whitespace, and characters outside the XML Char production in content, each
a located ParseError. The attribute scanners move to EventReaderAttributes
to stay under the type-body cap.
Add the opt-in W3C xmltest runner (XMLTS_ROOT-gated; the suite is never
vendored, James Clark's license permits redistribution only as the unmodified
archive; Foundation is used in this one test file for file access only).
Baseline: valid/sa 118/119, not-wf/sa 93/186 rejected (up from 80), the
remaining 93 carried as an exact known-deviations baseline so progress and
regressions are both visible. The two remaining leniency classes (internal-
subset DTD grammar, entity replacement-text well-formedness) are #120.
Part of #105 (Tier 2)
