leptris 1.9.201.0-arm-linux → 1.9.201.2-arm-linux

data/vendor-src/libleptris/TODO.xslt-full/REMAINING.md

@@ -0,0 +1,163 @@
+# REMAINING — the complete open-work ledger
+
+SSOT for everything still open after v1.9.144 (2026-09-12).
+Update entries as work ships; delete entries as they close.
+Per-lane details live in the numbered docs; this file is the index
++ the graph.
+
+## A. In flight — the staged path (2026-09-12)
+
+**STAGE 0 (shipping now)** — template fence slice on
+feat/html-template: WHATWG-gated in-template end-tag fence +
+structural html/head/body drop (attrs NOT merged) + template
+marker-clear in the AAFE list (corpus 1169 → 1175, parity 784
+held). The v1/v2 content-level heuristics were built, measured
+net-negative, and REVERTED — the remaining ~25 template reds need
+the real mode stack (14-html-mode.md entry (h) banked,
+entry (i) shipped; the mode-stack requirement lives there).
+
+**STAGE 1 — HTML tail (#659), the only engine lane owned
+end-to-end:**
+1.1 template insertion-mode stack: each open template saves its
+    insertion mode; starts switch it (tr → in_row, td → in_cell,
+    thead/tbody/caption/col → in_table), end tags restore; clears
+    the ~25 template.dat reds (floor → ~1200)
+1.2 the 3-red walker mystery (plain-text-unsafe 12/13,
+    noscript01:1) — engine output is correct, the corpus walker
+    compares the comment as ''
+1.3 remaining corpus families in size order (tests1/16 tails)
+1.4 bindings expose leptris_parse_html_string/html4
+    (leptris-ruby PR: Leptris::HTML.parse WHATWG + parse4
+    Nokogiri-parity) → CLOSE #659
+
+**STAGE 2 — XSLT/XPath/XQuery completion (lanes 05/07/11/12/15):**
+2.1 QT3 growth: UCA collation → error-assertion channel →
+    assert-type/assert-xml/any-of → prod/ + Context* sets → 3.1
+    sets (array/, map/)
+2.2 lane 05 dates tail: current-*, implicit-timezone,
+    adjust-*-to-timezone (fixed-offset model), format-date/
+    time/dateTime picture subset
+2.3 lane 07 fn-items tail: fn:sort key/collation arity,
+    for-each-pair past the zip
+2.4 lane 15: leptris_xpath_eval_versioned (1.0-strict vs 3.x
+    surface gating) + result-type mirrors in both bindings
+    (gate: abi spec + mirror-drift check)
+
+**STAGE 3 — Schematron lane 16: COMPLETE on the C side** (schema
+IR + SVRL evaluator + corpus 50/50 + `leptris validate`,
+v1.9.128/129). CLOSES when the user merges leptris-ruby#170 +
+leptris-py#99.
+
+**STAGE 4 — user-side gate (blocks #878 + lane 16/17 close):**
+merge leptris-ruby#165 (RelaxNG → then the standoc migration off
+Jing, prompt on #878 → #878 CLOSED), #170 (Schematron), #171
+(Diff), leptris-py#99 (Schematron), lutaml/canon#191 (structural
+fast path). All built, tested, waiting on the merge button.
+
+**STAGE 5 — #682 scope call (the only open decision):**
+(a) one more streaming phase (~+15%, lands ~1.1x in-process),
+(b) rescope the dispatch bar to >=1x in-process + >=2x xsltproc
+    wall (already true), (c) close as-is. Every user-visible row
+    is already >=2x. Do NOT micro-grind awaiting the call.
+
+**STAGE 6 — website benchmark-system v2:** owned by the website
+agent (prompt delivered 2026-09-12): XML/XSLT/YAML (@yeptris) x
+ruby/python/c scoreboard. We feed measured numbers only.
+
+## B. Open GitHub issues (leptris/leptris)
+
+- **#878** (nearly closed) — native RELAX NG validation. SHIPPED:
+  pattern IR + validator + backtracking fold (v1.9.109-111), Jing
+  conformance corpus gate 38/38 (v1.9.112), `<include>` via
+  leptris_rng_parse_file (v1.9.113), `<param>` facets (v1.9.114),
+  schema errors on leptris_last_error (v1.9.115), Ruby binding
+  (PR leptris-ruby#165). REMAINING: merge #165 → standoc
+  migration off Jing (drafted on the issue) → close.
+- **#682** (banked, USER SCOPE CALL pending) — in-process dispatch
+  heavy 0.80x / light 0.76x vs lxml after the shipped levers
+  (hash+mode-bucket dispatch, TLS consolidation, walk-first attr
+  dup, AVT fast path, AST-cache TLS claim, streaming phase 1).
+  Ceiling analysis (TODO 13): streaming remainder (~15%) + eval
+  rewrites (~10%) + micro (~5-8%) tops out ~1.1x in-process —
+  the >=2x bar is unreachable there while libxslt does identical
+  work. Bars HELD: transform 5x, predicate 15.6x, value-of 135x,
+  subtree-copy 2.27x, xsltproc wall >=2x. Options on the issue.
+- **#659** — v1.9.116-119 jumps, then the v1.9.137-144 WHATWG
+  arc: full AFE list + foreign scope boundaries (933 → 1027),
+  in-table clear-stack + frameset (→ 1038), bogus-markup edges
+  (→ 1046), script-data escaped states (→ 1096), RCDATA family
+  (→ 1129), numeric-reference end states — entities01 100% green
+  (→ 1169), template fence subset (→ 1175 of 1753). Parity 784
+  held throughout; html4 mode stays byte-pinned. REMAINING =
+  Stage 1 above. (Also shipped: leptris_element_expanded_name,
+  v1.9.144, the adapter batched-accessor primitive.)
+- **#930** — CLOSED (v1.9.117).
+
+## C. TODO.xslt-full lanes — residual items
+
+- 01/03/04/06/08/09/10 — CLOSED.
+- 05 (dates) tail → Stage 2.2. 07 (fn items) tail → Stage 2.3.
+- 11/12 (XQuery) tails → Stage 2.1 QT3 subsets via the XTH
+  connector shape.
+- 13 (perf) = #682 (Stage 5). SAX binding-drain residual: retry
+  quiet-machine best-of-20, else engine-side batched-drain API.
+- 14 (HTML) = #659 (Stage 1).
+- 15 (binding entries) → Stage 2.4. RelaxNG entries already
+  shipped (leptris-ruby#165).
+- 16 (Schematron 2025): COMPLETE C-side = Stage 3; binding PRs
+  open (user merges).
+- 17 (XML diff): PRs open = Stage 4 (leptris-ruby#171, canon#191).
+
+## D. Bindings (user-owned repos — PRs only, never release)
+
+- leptris-ruby#165 (RelaxNG), #170 (Schematron), #171 (Diff),
+  leptris-py#99 (Schematron) — all open, ready, user merges.
+- moxml perf rows: parse 2.6-2.8x, serialize 2.4x, e2e 2.15x,
+  xpath 31x, HTML 3.6-11x, C14N 23x AHEAD of Nokogiri; the 0.69x
+  NS-read row CLOSED 2026-09-12 (four measurements — the deficit
+  is moxml's own materializer layer; our C resolver and gem are
+  measured clear).
+
+## E. External (Jing / metanorma-pdfa#98 follow-ups)
+
+- sshaw/ruby-jing#6 upstream dormant; the strategic fix (#878) is
+  shipped — the standoc migration is the remaining step.
+- metanorma/standoc#1244 + mnconvert-ruby#40 — user review/merge.
+
+## F. Housekeeping
+
+- ~20 stale local build* dirs (user's call; one-build-dir
+  discipline going forward — Mac crash memory).
+- Pre-existing warnings on main: html_parse.c:2160 (dual anonymous
+  entity-table struct types) + sax/pull.c:447 (const discard).
+  Both benign; fix in a cleanup PR.
+- Known repo quirk: libxslt-suite fixtures committed with CRLF
+  bytes while .gitattributes says text eol=lf (e.g. bug-80.xml)
+  — any stat-cache invalidation shows them dirty with an
+  EOL-only diff. Do NOT "fix" by renormalizing inside an
+  unrelated PR.
+
+## Graph to success
+
+```
+STAGE 0 template fence slice (feat/html-template) → v1.9.145
+  │
+STAGE 1 #659: template mode stack → walker reds → family tails
+  → bindings html entries → #659 CLOSED
+  │
+STAGE 2 lanes 05/07/11/12/15: QT3 + dates + fn-items + versioned
+  eval entries → language surface complete, suite-gated
+  │
+STAGE 3+4 user merges: ruby#165 → standoc off Jing → #878 CLOSED;
+  #170 + py#99 → lane 16 CLOSED; #171 + canon#191 → lane 17 CLOSED
+  │
+STAGE 5 #682 scope call → lane 13 closed
+  │
+STAGE 6 website v2 (agent) fed by our measured scorecard
+```
+
+XSLT speed vs the field (current, gated): transform 5x, predicate
+15.6x, value-of 135x, subtree-copy 2.27x, xsltproc wall >=2x,
+HTML parse 2.9x, C14N 23x binding-armed — every USER-VISIBLE
+feature is ahead; the only sub-parity row is in-process dispatch
+(0.80x, ceiling-analyzed, awaiting the user's scope call #682).

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/00-triage.md

@@ -0,0 +1,171 @@
+# 00 — Triage board (updated 2026-08-26, suite 79/205)
+
+Live census of test/xslt/open_cases.txt after the root-semantics,
+namespace-axis, attr-set-chain, and serializer batches:
+- ~92 DIFF (mixed: number formatting, sort, document(), strip-space
+  corners, AVT quirks, stale-fixture bug-124)
+- ~17 NUMBER, ~9 SORT, ~8 KEY, ~7 DOCUMENT, ~6 EMPTY/crash
+- Windows-only divergences gated on the list: bug-93, bug-102
+  (xsl:import resolution on Win32), bug-118 (copy-of select='/'
+  document copy). Fix the Win32 path/join in the import resolver and
+  the document-copy path, then un-gate.
+
+# 00 — Triage harness (standing tooling)
+
+The open list is living state, not a snapshot. Ship the triage as a
+script (scripts/xslt_triage.sh) so every PR regenerates it:
+- Runs each open case via the fork-isolated runner.
+- Buckets by signature: EMPTY / CRASH / NS-OUT / CHARREF / PI-OUT /
+  DIFF with one-line diffs.
+- Rewrites test/xslt/open_cases.txt; CI fails if a case in the list
+  passes or a case outside it fails (already enforced by the suite
+  runner — this script keeps the list honest).
+
+Done: the ad-hoc python used so far; productize it (SSOT: one
+script, used by CI + devs).
+
+---
+## Session log 2026-08-26 (72/205)
+
+Landed: cdata-section-elements + merged CDATA runs; prefix/ns-preserving
+copies; top-level comment/PI materialization (before/after root chains);
+xsl:number any-level node-kind walk; xsl:comment/PI content capture;
+HTML PI serialization; RTF variables bind the fragment document node
+(fragment chains = doc children on all three axis paths); namespace-gated
+extension-function local-name fallback; attribute nodes in template
+matching (@-patterns, copy, unified node-kind matcher); control-char
+attr escaping on the mutation path.
+
+Remaining census: ~85 DIFF, ~15 NUMBER, 9 SORT, 8 KEY, 7 DOCUMENT,
+~5 crash. Windows gates: bug-93, bug-102 (import path), bug-118.
+
+---
+## Session log 2 (79/205, HEAD 72d2974)
+
+Landed after the morning batch: attribute nodes in template matching
+(@-patterns + unified node-kind matcher used by BOTH the match and the
+priority rescan); xsl:copy of attributes; control-char attr escaping on
+the mutation serializer path; §2.4 stylesheet whitespace (strip ALL
+ws-only text — LRE indentation does NOT survive; xsl:text and
+xml:space=preserve excepted — worth +4); in-scope (ancestor-walked)
+namespace copies with redundant-declaration suppression (bug-128);
+self::node() from attribute nodes; ancestor-or-self reverse document
+order; document() → document node (bug-153) + foreign doc nodes resolve
+their OWN root (bug-29- inclusion-loop crash); RTF variables bind the
+fragment document node (bug-112/143/144); namespace-gated extension
+local-name fallback.
+
+Next clusters: ~80 DIFF (mixed one-offs — AVT-in-attribute-name
+{name()} cases bug-195/196/197, format-number grouping bug-75/95,
+xsl:number NaN→0 quirk bug-187), 16 NUMBER, 8 SORT, 7 KEY, ~5 crash.
+Windows gates: bug-93, bug-102 (xsl:import path resolution on Win32),
+bug-118 (copy-of select='/').
+
+---
+## Session log 3 (102 → 129/205, HEAD after 934acb9 + 5 commits)
+
+Reference-oracle upgrade: libxml2 2.16.0 + libxslt checkouts at
+~/src/external — all serializer behavior now ported from source
+(HTMLtree.c htmlNodeDumpInternal, xsltutils.c xsltSaveResultTo),
+not empirical probing.
+
+Landed:
+1. libxml2-parity HTML output (e709dad): serializer html_method
+   mode — element descriptor table (EMPTY/INLINE/BLOCK, bsearch),
+   the four newline sites, zero nesting spaces, p/pre/param
+   parent-side rule; DOM-level <meta charset> injection
+   (htmlSetMetaEncoding port); §16.1 unnamespaced-html-root default
+   method + indent tri-state (-1: html yes, xml no); non-void empty
+   → open+close pairs; full void-element set.
+2. position() (2cb8cc5): exec carries current_pos; for-each +
+   apply-templates (selection + default child walk) set it; new
+   leptris_xpath_compiled_eval_ctx seeds ctx->context_position
+   (VM + interpreter read the same field).
+3. strip-space '*' (934acb9): NameTest wildcard in name_in_list;
+   zero-length text children are stripped nodes (parse never
+   creates them) — skipped in the child walk.
+4. call-template §11 scope (later commit): exec snapshots the
+   globals-only chain head; call resets there.
+5. copy-of verbatim mixed content: every child kind in document
+   order (was string-value + elements only).
+6. Element-less results keep the XML declaration (§16.1).
+7. EXSLT func module (3 commits): func:function/func:result/
+   func:param — registry bindings, globals-only scope, RTF
+   node-set returns, caller-context arg binding, function-ns
+   output exclusion (pre-scan before template compile).
+8. Namespace-URI pattern matching: XsltTemplate.ns (build_ns_
+   context), pattern ladder evaluates through it; selection-loop
+   temp template keeps .ns; /name fast path resolves the test
+   prefix through ctx->ns_set. Cross-prefix matches now correct.
+
+Suite: 129/205. Remaining census (76): 35 no-marker one-offs,
+7 xsl:number, 7 namespace, 6 document(), 5 import, 4 id(), 3 key,
+2 format-number, 2 exsl, ~10 more. Named next targets:
+- bug-53: DTD ATTLIST DEFAULT attribute materialization at parse
+  (parser-level feature — the dtd/ model stores default_value but
+  nothing applies it on the parse path)
+- bug-83: HTML URL-escaping of href attribute values (%20)
+- bug-33-: script/style raw text (serializer isRaw/dataMode)
+- bug-216: post-root document text (source trailing \n)
+- Windows gates: bug-93, bug-102, bug-118 — now PASSING on POSIX,
+  removed from the worklist (runner is POSIX-forked)
+
+---
+## Session log 4 (129 → 131/205 + architecture/perf pass)
+
+Architecture review (report: architecture-review-20260826-xslt-perf.html,
+4 candidates) — ALL IMPLEMENTED:
+
+1. exec-scoped eval environment: document-cached function registry
+   (invalidate on custom-fn register / exslt_enable / transform
+   enter-exit) + dirty-skip varset mirror. Transform perf:
+   bug-5- 115.3 → 48.5 ms (2.4x), doc 0.204 → 0.056 ms (3.6x).
+2. HTML single-owner in the serializer: voids, non-void pairs,
+   attr escaping (raw ', href/src URI %XX), rawtext elements
+   (script/style/...), PI '>' close, indent=no semantics, method
+   tri-state. to_html_method + html_post_pass DELETED (~200 lines).
+   Suite: bug-83 minimal shape exact.
+3. position() SSOT folded into leptris_xpath_compiled_eval_ctx.
+4. Recursive serializer walker DELETED (-369/+64): heap-growable
+   frames (2000-deep verified, ASAN-clean).
+
+Latent bugs fixed along the way (ASAN-exposed):
+- xslt_capture_children_text: stale rtf_text_cap overflowed the
+  fresh buffer on nested captures
+- buffer_create: uninitialized cdata_names/cdata_count read as the
+  cdata list in every options-NULL serialization
+- text-node base.raw never initialized in either creator
+- leptris_node_prepend_child SEVERED the old first child's sibling
+  link — every prepend dropped the children-chain tail (public API
+  bug, exposed by the meta injection)
+
+Known-environmental: test_cli spawns die with SIGKILL on this
+machine (binary runs clean from shell; reproduced on HEAD stashed).
+
+Suite 131/205; 983/983 ctest green. Remaining 74: same census as
+log 3 minus closed items (bug-33-/83/159 cluster partially closed).
+
+---
+## Session log 5 (136/205, morning 2026-08-27)
+
+Landed: §11 scoping moved into xslt_invoke_template (apply-templates
+twin, bug-42-); ATTLIST default-value materialization at parse
+(bug-53/90 cluster); strip/preserve-space prefixed NameTests p:* (bug-124);
+UTF-8 NCNames in the XPath lexer (bug-155); leading div/mod/and/or
+tokens are NameTests — match="div" / "div|obj" compile (bug-70);
+serializer ns order (default first, bug-150/155); xsl:output encoding
+in the result declaration (bug-132).
+
+KNOWN OPEN with root cause (bug-16-): `//NAME` and
+`/descendant::NAME` from a document-node ("/") context DROP the root
+element — the absolute-path branch seeds the walk with the ROOT
+ELEMENT, so descendant:: walks children-only from it. Seeding with
+the DOCUMENT node instead fixes /descendant:: but over-counts the
+-or-self forms (double root add) and still misses //NAME's child
+step from the doc node — the proper fix needs the doc-branch
+child/descendant routes audited together (attempt reverted; the
+regression risk was not worth the single case at session end).
+count(//*), /descendant-or-self::*, /NAME unaffected.
+
+Suite 136/205. Remaining census: ~31 no-marker one-offs, 7 number,
+7 namespace, 6 document(), 5 import, 3 id, 3 key.

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/01-fragment-nodes.md

@@ -0,0 +1,19 @@
+# 01 — Fragment-level PIs, comments, and text order (84 cases)
+
+The largest cluster: literal PIs/comments in template content and
+fragment-level output misorder or drop. libxslt keeps exact
+document order of top-level result nodes (elements, PIs, comments,
+text interleaved).
+
+Work:
+- The result FRAGMENT is a node sequence, not "elements + two text
+  buffers". Replace top_text/tail_text/frag_nodes with one ordered
+  fragment chain on XsltExec (MECE: exactly one representation).
+- out_append_* routes every node kind through it; serialization
+  walks it once (DRY with the serializer's doc-order walk).
+- xsl:processing-instruction / xsl:comment / xsl:text at fragment
+  level then serialize in order (bug-107, bug-100 family, bug-11-).
+
+Verify: xsltproc on each case first; then the suite delta.
+Perf gate: the fragment chain replaces THREE buffers — allocation
+count must drop on benchmarks/xslt/fragment.

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/02-root-node-semantics.md

@@ -0,0 +1,34 @@
+# 02 — Document-node context for match='/' — DONE (2026-08-25, a524706)
+
+# 02 — Document-node context for match='/' (part of EMPTY, ~12 cases)
+
+match='/' is the DOCUMENT node: child::*/node() from it select the
+root element; count(node()) = 1 even with comments/PIs before it.
+Our engine conflates "/" with the root element.
+
+Work (SSOT — fix at the model, not per-expression):
+- Give the evaluator a real document-node context entry point: a
+  context flag + child/descendant axis handlers that treat the doc
+  root as the child of the virtual document node (ONE site in
+  evaluate_step's axis dispatch — not expression-name intercepts;
+  the intercept attempt crashed and was reverted).
+- Root invocation sets the doc context; apply-templates default
+  enumerates [root element]; built-in rules unchanged.
+- Covers bug-104 (copy-of select='*'), bug-119, bug-124 family.
+
+Verify: name(/*), count(node()), select='*' from "/" all agree
+with xsltproc; 45/45 XsltFull stays green (match='/r' is ELEMENT
+context — distinct semantics, one test each way).
+
+---
+## DONE — implemented in a524706
+
+- DOM: LEPTRIS_NODE_TYPE_DOCUMENT singleton (dom/document_node.{h,c}).
+- Evaluator: doc-node child/descendant axes (VM + AST paths); self/parent
+  from non-element nodes.
+- XSLT: entry + built-in root rule run on the document node; "/" matches
+  only it; apply-templates/copy/copy-of take the node-typed accessor.
+- Identity transforms (bug-107 family) verified end-to-end.
+- Remaining known open: bug-119 (libxslt emits the literal body of a
+  source-side xsl:template matched via an attribute pattern — a
+  recorded-libxslt-quirk case; needs a quirk-equivalence decision).

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/03-crashes.md

@@ -0,0 +1,15 @@
+# 03 — Crash cases (10): depth accounting + cycle hardening
+
+bug-6-, bug-70, bug-100, bug-111, bug-130, bug-140, bug-142,
+bug-147, bug-155, bug-157 crash the process.
+
+Work:
+- XsltExec depth counter (not a stack-canary gamble): increment in
+  xslt_exec_instrs AND xslt_invoke_template; at XSLT_MAX_DEPTH
+  return a recoverable error (per §14 terminations), never SIGSEGV.
+- Include/import cycle guard exists; add document() cycle guard
+  (same chain table) — bug-130 imports four .imp files deep.
+- Each crasher gets its own root-cause line in the PR (10 lines);
+  if a crash is an XPath-layer bug (bug-100), file it separately
+  and cross-link.
+Perf gate: the counter is one increment/return — zero measurable.

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/04-namespaces.md

@@ -0,0 +1,14 @@
+# 04 — Namespace output corners (9 NS-OUT + parts of PI-OUT)
+
+Namespaced results lose declarations or drop them on the floor.
+
+Work:
+- xmlns="" (unset inherited default) — build_ns_copy must carry an
+  "empty default" state and emit it (bug-103).
+- Prefixed xsl:attribute namespace= → declaration lands on the
+  PARENT element (bug-122, bug-104).
+- copy-of of namespaced trees copies in-scope decls onto each copy
+  (§7.5) — copy_node_deep walks the SOURCE element's decl list.
+- xsl:namespace-alias + prefixed LRE output (bug-129).
+- xml:base/xml:lang/xml:space should pass through (bug-12- cluster).
+Verify against xsltproc; suite delta.

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/05-number-attrset-sort.md

@@ -0,0 +1,13 @@
+# 05 — xsl:number, attribute-set precedence, sort/AVT corners (24 DIFF)
+
+- xsl:number: `from` stopping semantics at depth, level=any across
+  comment/PI interleaving, format="⑴"-style single-token
+  corners, lang (bug-25-, bug-63, bug-90 family).
+- xsl:attribute-set: LAST declaration wins across imports (our
+  lookup takes first); use-attribute-sets chains resolve at USE
+  time with import rank (bug-93, bug-102 family).
+- xsl:sort: data-type="number" with leading/trailing whitespace
+  keys; lang collation falls back to codepoint ONLY when lang
+  unsupported (bug-120 family).
+- AVT: `{{`/`}}` escaping corners; `{$x` literal (bug-126).
+All SSOT: one comparator, one priority resolver, one AVT scanner.

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/06-document-chains.md

@@ -0,0 +1,10 @@
+# 06 — document() multi-file + remaining EMPTY (~16)
+
+- document('file.xml', node) second-arg base resolution (same
+  directory as the stylesheet — the suite's hrefs are relative).
+- Cached docs share the rtf_chain lifetime model; cross-document
+  nodes serialize with their OWN ns decls.
+- id()/key() against document('') trees (bug-115, bug-160).
+- Early-exit audit: any op_ handler returning -1 kills output —
+  replace silent aborts with recoverable errors (MECE: recovery
+  policy lives in ONE place — the walker).

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/07-charef-serializer.md

@@ -0,0 +1,9 @@
+# 07 — Serializer char corners (2 CHARREF + decl/indent polish)
+
+- bug-74: newline+tab in attribute values → 
	 (numeric
+  refs, not literal bytes) — the mutation-attr path bypasses the
+  inline emitter I patched; route BOTH attr emitters through one
+  escape table (SSOT — one table, two entry points).
+- bug-159: HTML method percent-encoding + charset meta emission.
+- XML decl: emit only when input had one OR method=xml with
+  standalone; exact attribute order version/encoding/standalone.

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/08-perf-gate.md

@@ -0,0 +1,8 @@
+# 08 — Performance gate + competitor matrix (standing)
+
+Benchmarks (benchmarks/xslt/): fragment-heavy, namespace-heavy,
+number-heavy, key-lookup, document()-chain — one per fixed cluster.
+Matrix: leptris vs xsltproc vs Saxon-HE (java -jar saxon-he.jar)
+on the same inputs; CI job runs it and FAILS on >3% regression
+vs our own baseline AND logs the gap vs competitors. Report table
+in the PR (like the XPath one on the website).

data/vendor-src/libleptris/TODO.xslt-full/legacy-v19/09-release.md

@@ -0,0 +1,14 @@
+# 09 — v1.10.0 release (gated)
+
+Preconditions (ALL):
+- test/xslt/open_cases.txt EMPTY (205/205 libxslt suite)
+- 45/45 XsltFull, 438/438 XPath conformance, full ctest green
+  serially AND parallel (allowing only the documented macOS CLI
+  spawn flake)
+- ASAN (Linux) clean, leaks (macOS) clean, TSAN clean, Windows
+  builds green
+- Benchmarks: no >3% regression vs main; competitor table attached
+
+Then: minor bump via the automated workflow
+(gh workflow run release.yml -f next_version=1.10.0), real release
+notes in the release PR, rebase-merge, verify tag + GitHub Release.

data/vendor-src/libleptris/VALIDATION.md

@@ -0,0 +1,151 @@
+# Leptris Validation Commands
+
+## Quick Validation (Run all tests and benchmarks)
+
+```bash
+# Run the validation script
+./scripts/validate.sh
+```
+
+## Individual Test Commands
+
+### Run all C tests
+```bash
+cd build
+ctest --test-dir build --output-on-failure
+```
+
+### Run specific test suites
+```bash
+# DOM tests
+./build/test/c/test_dom
+
+# XPath tests
+./build/test/xpath/test_xpath
+
+# Parser tests
+./build/test/test_parse
+
+# CLI tests
+./build/test/cli/test_cli_commands
+```
+
+### Run benchmarks
+```bash
+# DOM benchmark
+./build/benchmarks/dom_benchmark benchmarks/fixtures/small.xml 1000
+
+# DOM modify benchmark
+./build/benchmarks/bench_dom_pugixml
+
+# DOM benchmark v2 (parse once, measure operations)
+./build/benchmarks/dom_benchmark_v2
+```
+
+## Memory Leak Detection
+
+### macOS
+```bash
+leaks --atExit -- ./build/test/c/test_dom
+```
+
+### Linux (valgrind)
+```bash
+valgrind --leak-check=full --error-exitcode=1 ./build/test/c/test_dom
+```
+
+## CLI Testing
+
+```bash
+# Parse and display XML
+./build/cli/leptris parse test/fixtures/libxml2/svg1
+
+# XPath query
+./build/cli/leptris xpath test/fixtures/libxml2/svg1 "//svg"
+
+# Format XML
+./build/cli/leptris format --indent 2 test/fixtures/libxml2/svg1
+```
+
+## Build Commands
+
+### Minimal build (no optional features)
+```bash
+cmake -B build -S . \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DLEPTRIS_BUILD_CLI=ON \
+    -DLEPTRIS_ENABLE_UTF8PROC=OFF \
+    -DLEPTRIS_ENABLE_ICONV=OFF
+cmake --build build
+```
+
+### Full build with all features
+```bash
+cmake -B build -S . \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DLEPTRIS_BUILD_CLI=ON \
+    -DLEPTRIS_ENABLE_UTF8PROC=ON \
+    -DLEPTRIS_ENABLE_ICONV=ON \
+    -DLEPTRIS_BUILD_BENCHMARKS=ON \
+    -DBUILD_TESTING=ON
+cmake --build build
+```
+
+### Build with vcpkg dependencies
+```bash
+cmake -B build -S . \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DCMAKE_TOOLCHAIN_FILE=$VCPKG_ROOT/scripts/buildsystems/vcpkg.cmake \
+    -DLEPTRIS_BUILD_CLI=ON
+cmake --build build
+```
+
+## Expected Results
+
+### Test Coverage
+- XPath W3C Conformance: 438/438 tests (100%)
+- CLI Tests: 88/88 tests (100%)
+- DOM Tests: 105/106 tests (99.1%)
+- Total: 777+ tests
+
+### Performance Targets
+- XPath vs libxml2: ≥1.5x faster ✅
+- DOM parsing: Competitive with pugixml (target: ≥1.2x)
+- Memory: Compact element structure (~96 bytes vs 152+ bytes in legacy design)
+
+## Troubleshooting
+
+### Build failures
+```bash
+# Clean build directory
+rm -rf build
+cmake -B build -S . [options]
+```
+
+### Test failures
+```bash
+# Run tests with verbose output
+ctest --test-dir build --output-on-failure --verbose
+
+# Run individual test for detailed output
+./build/test/c/test_dom --gtest_filter=TestName.TestCase
+```
+
+### Benchmark failures
+
+Benchmarks require pugixml for comparison. Install it first:
+
+```bash
+# Install pugixml (macOS)
+brew install pugixml
+
+# Install pugixml (Ubuntu/Debian)
+sudo apt-get install libpugixml-dev
+
+# Or build from source
+git clone https://github.com/zeux/pugixml.git
+cd pugixml
+cmake -B build -S .
+cmake --build build
+sudo cmake --install build
+```