Release v0.7.0

CHANGELOG.md

@@ -6,18 +6,31 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+## [0.7.0] - 2026-03-14
+
 ### Added
 - Unit tests for `blocks.lua`, `academic.lua`, `reader_inlines.lua`, `reader_blocks.lua`, `reader_academic.lua` (235 new assertions, 505 total across 11 test files)
 - Golden baseline comparison (`make test-golden`) in CI pipeline
 - `make test-all` target for single-command full validation (lint + test + golden + reader + validate)
 - Wildcard test discovery in `make test-unit` (auto-discovers new test files)
+- Unit tests for task list checkboxes, subfigure extraction, LineBlock handler, inline edge cases (Quoted, SmallCaps, Span anchors), and reader block handler output (130 new assertions, 621 total)
+- Reader round-trip golden baselines (`make test-reader-golden`, `make update-reader-golden`)
+- Shared `THEOREM_VARIANTS` and `THEOREM_REF_PREFIXES` constants in `lib/utils.lua`
+- SECURITY.md with supported version policy
 
 ### Changed
+- **Content version** bumped from `0.7.0` to `0.7.1` to align with cdx-core v0.7.1
 - Aligned `make lint` flags with CI (`--no-unused-args --no-max-line-length`)
 - Updated CONTRIBUTING.md pre-PR command to `make test-all`
+- Consolidated theorem variant and reference prefix definitions from `academic.lua`/`inlines.lua` into shared `utils.lua` constants
+- Removed dead code: unused `caption` parameter from `blocks.image()`, unused `convert_inline`/`_handlers` exports from `inlines.lua`, `has_class` re-export from `inlines.lua`
 
 ### Fixed
 - README Pandoc version requirement: corrected from 2.11+ to 3.0+
+- Table cell complex content: now collects all child nodes from nested blocks, not just text nodes
+- Measurement sentinel: defaults to `value=0`/`unit=""` with stderr warnings for unparseable input
+- RawBlock format preservation: sets `language` field from block format (e.g., `html`, `latex`)
+- Citation prefix/suffix: prefix applied only to first citation, suffix/locator only to last (multi-citation fix)
 
 ## [0.6.0] - 2026-02-17
 

Makefile

@@ -9,7 +9,7 @@ TEST_INPUTS := $(wildcard tests/inputs/*.md)
 TEST_OUTPUTS := $(patsubst tests/inputs/%.md,tests/outputs/%.json,$(TEST_INPUTS))
 TEST_CDX := $(patsubst tests/inputs/%.md,tests/outputs/%.cdx,$(TEST_INPUTS))
 
-.PHONY: all test clean test-json test-cdx test-reader test-unit test-golden test-all help check-deps validate-schema lint
+.PHONY: all test clean test-json test-cdx test-reader test-unit test-golden test-reader-golden test-all help check-deps validate-schema lint
 
 all: test
 
@@ -139,6 +139,31 @@ update-golden: test-json
 	done
 	@echo "Golden baselines updated."
 
+# Compare reader round-trip outputs against golden baselines
+# Note: Pandoc Span/Div attribute ordering is non-deterministic between process
+# invocations, so this target is NOT included in test-all. Use update-reader-golden
+# to capture baselines, then run test-reader-golden in the same make invocation:
+#   make update-reader-golden && make test-reader-golden
+test-reader-golden: test-reader
+	@echo "Comparing reader round-trip against golden outputs..."
+	@fail=0; for f in tests/outputs/*.roundtrip.md; do \
+		base=$$(basename $$f); \
+		if [ -f tests/expected/$$base ]; then \
+			diff -q $$f tests/expected/$$base > /dev/null 2>&1 || { echo "DIFF: $$base"; diff tests/expected/$$base $$f | head -20; fail=1; }; \
+		fi; \
+	done; \
+	[ $$fail -eq 0 ] && echo "All reader golden tests passed." || { echo "Reader golden test failures detected."; exit 1; }
+
+# Regenerate reader round-trip golden baselines
+update-reader-golden: test-reader
+	@echo "Updating reader round-trip golden baselines..."
+	@mkdir -p tests/expected
+	@for f in tests/outputs/*.roundtrip.md; do \
+		base=$$(basename $$f); \
+		cp $$f tests/expected/$$base; \
+	done
+	@echo "Reader golden baselines updated."
+
 # Run all tests (unit + integration + golden + reader + lint + validate)
 test-all: lint test test-golden test-reader validate
 

README.md

@@ -238,7 +238,7 @@ The writer produces a JSON structure with three sections:
     }
   },
   "content": {
-    "version": "0.7.0",
+    "version": "0.7.1",
     "blocks": [...]
   },
   "dublin_core": {
@@ -380,6 +380,10 @@ The reader converts Codex back to standard Pandoc elements. Most block types sur
 - [codex-file-format-spec](https://github.com/Entrolution/codex-file-format-spec) - Format specification
 - [cdx-core](https://github.com/Entrolution/cdx-core) - Rust library and CLI
 
+## Security
+
+See [SECURITY.md](SECURITY.md) for the supported version policy and how to report vulnerabilities.
+
 ## Contributing
 
 Contributions are welcome! Please read [CONTRIBUTING.md](CONTRIBUTING.md) before submitting PRs.

SECURITY.md

@@ -0,0 +1,29 @@
+# Security Policy
+
+## Supported Versions
+
+| Version | Supported          |
+|---------|--------------------|
+| 0.7.x   | :white_check_mark: |
+| < 0.7.0 | :x:                |
+
+Only the latest patch release of the current minor version receives security updates.
+
+## Reporting a Vulnerability
+
+If you discover a security vulnerability, please report it responsibly:
+
+1. **Do not** open a public issue.
+2. Email the maintainers or use [GitHub Security Advisories](https://github.com/Entrolution/cdx-pandoc/security/advisories/new) to report privately.
+3. Include a description of the vulnerability, steps to reproduce, and any potential impact.
+
+We will acknowledge receipt within 48 hours and aim to provide a fix or mitigation plan within 7 days.
+
+## Scope
+
+cdx-pandoc is a document conversion tool that processes untrusted input (Markdown, LaTeX, Word, etc.) via Pandoc. Security considerations include:
+
+- **Input handling**: The Lua writer processes Pandoc AST structures. Malformed input is handled by Pandoc's parser before reaching this code.
+- **Shell script**: `scripts/pandoc-to-cdx.sh` invokes external tools (`pandoc`, `jq`, `sha256sum`, `zip`). File paths are quoted to prevent injection.
+- **No network access**: The writer and reader operate entirely offline with no network calls.
+- **No code execution**: The writer produces static JSON output. No user-supplied code is evaluated.

codex.lua

@@ -53,7 +53,7 @@ blocks.set_academic(academic)
 
 -- Spec version
 local CODEX_VERSION = "0.1"
-local CONTENT_VERSION = "0.7.0"
+local CONTENT_VERSION = "0.7.1"
 
 -- Generate ISO 8601 timestamp
 local function iso_timestamp()

lib/academic.lua

@@ -23,19 +23,21 @@ function M.set_extension_tracker(tracker)
     track_extension = tracker or function() end
 end
 
--- Theorem variant set
-local theorem_variants = {
-    theorem = true, lemma = true, proposition = true, corollary = true,
-    definition = true, conjecture = true, remark = true, example = true
-}
+-- Theorem variant set (from shared constants)
+local theorem_variants = utils.THEOREM_VARIANTS
 
 -- Academic class set (all classes handled by this module)
-local academic_classes = {
-    theorem = true, lemma = true, proposition = true, corollary = true,
-    definition = true, conjecture = true, remark = true, example = true,
-    proof = true, exercise = true, ["exercise-set"] = true,
-    algorithm = true, abstract = true, ["equation-group"] = true
-}
+-- Derived from theorem variants plus non-theorem academic types
+local academic_classes = {}
+for k in pairs(theorem_variants) do
+    academic_classes[k] = true
+end
+academic_classes.proof = true
+academic_classes.exercise = true
+academic_classes["exercise-set"] = true
+academic_classes.algorithm = true
+academic_classes.abstract = true
+academic_classes["equation-group"] = true
 
 -- Classify a Div as an academic block type or nil
 -- @param classes Array of CSS classes
@@ -363,12 +365,20 @@ function M.equation_group(text)
         return nil
     end
 
+    -- Split on \\ (double backslash) line breaks using plain string find
+    -- Cannot use gmatch character class here: [^\\\\]+ splits on every single
+    -- backslash, which would fragment LaTeX commands like \frac, \alpha, etc.
     local lines = {}
-    for line in inner:gmatch("[^\\\\]+") do
-        local trimmed = line:match("^%s*(.-)%s*$")
+    local pos = 1
+    while pos <= #inner do
+        local s, e = inner:find("\\\\", pos, true)
+        local segment = s and inner:sub(pos, s - 1) or inner:sub(pos)
+        local trimmed = segment:match("^%s*(.-)%s*$")
         if trimmed and trimmed ~= "" then
             table.insert(lines, trimmed)
         end
+        if not s then break end
+        pos = e + 1
     end
 
     return {

lib/blocks.lua

@@ -83,12 +83,16 @@ block_handlers.DefinitionList = function(block) return M.definition_list(block)
 block_handlers.Figure = function(block) return M.figure(block) end
 
 block_handlers.RawBlock = function(block)
-    return {
+    local result = {
         type = "codeBlock",
         children = {
             {type = "text", value = block.text}
         }
     }
+    if block.format and block.format ~= "" then
+        result.language = block.format
+    end
+    return result
 end
 
 block_handlers.LineBlock = function(block)
@@ -517,13 +521,11 @@ function M.table_cell(cell)
         if #blocks > 0 and blocks[1].type == "paragraph" then
             children = blocks[1].children
         else
-            -- Complex cell content - just use first text we find
+            -- Complex cell content - collect all children from converted blocks
             for _, b in ipairs(blocks) do
                 if b.children then
                     for _, c in ipairs(b.children) do
-                        if c.type == "text" then
-                            table.insert(children, c)
-                        end
+                        table.insert(children, c)
                     end
                 end
             end
@@ -807,7 +809,7 @@ function M.extract_subfigure(div, attrs, div_id)
 end
 
 -- Convert Image inline to image block (with optional dimensions)
-function M.image(img, caption)
+function M.image(img)
     local src = img.src or img.target or ""
     local alt = ""
 
@@ -840,14 +842,6 @@ function M.image(img, caption)
         end
     end
 
-    -- Caption passed from outside (legacy path)
-    if caption and caption.long then
-        local cap_text = pandoc.utils.stringify(caption.long)
-        if cap_text and cap_text ~= "" then
-            result.title = cap_text
-        end
-    end
-
     return result
 end
 

lib/inlines.lua

@@ -4,6 +4,7 @@
 -- Load shared utilities
 local utils = dofile((PANDOC_SCRIPT_FILE and (PANDOC_SCRIPT_FILE:match("(.*/)" ) or "") or "") .. "lib/utils.lua")
 local deep_copy = utils.deep_copy
+local has_class = utils.has_class
 
 local M = {}
 
@@ -37,9 +38,6 @@ function M.reset_context()
     M._default_context = M.new_context()
 end
 
--- Check if a class list contains a specific class (delegates to utils)
-M.has_class = utils.has_class
-
 -- Check if two marks are equal
 local function marks_equal(m1, m2)
     if type(m1) ~= type(m2) then
@@ -199,11 +197,8 @@ inline_handlers.Code = function(inline, marks, ctx)
     return {M.text_node(inline.text, new_marks)}
 end
 
--- Academic cross-reference prefix patterns
-local theorem_ref_prefixes = {
-    ["thm-"] = true, ["lem-"] = true, ["prop-"] = true, ["cor-"] = true,
-    ["def-"] = true, ["conj-"] = true, ["rem-"] = true, ["ex-"] = true
-}
+-- Academic cross-reference prefix patterns (from shared constants)
+local theorem_ref_prefixes = utils.THEOREM_REF_PREFIXES
 
 local function detect_academic_ref(target)
     if not target or not target:match("^#") then
@@ -358,7 +353,7 @@ inline_handlers.Span = function(inline, marks, ctx)
     local has_semantic_mark = false
 
     -- Entity class
-    if M.has_class(classes, "entity") then
+    if has_class(classes, "entity") then
         track_extension(utils.EXT_SEMANTIC)
         local entity_mark = {type = "semantic:entity"}
         if attributes.uri then
@@ -377,7 +372,7 @@ inline_handlers.Span = function(inline, marks, ctx)
     end
 
     -- Glossary class
-    if M.has_class(classes, "glossary") then
+    if has_class(classes, "glossary") then
         track_extension(utils.EXT_SEMANTIC)
         local glossary_mark = {type = "semantic:glossary"}
         if attributes.ref then
@@ -391,10 +386,18 @@ inline_handlers.Span = function(inline, marks, ctx)
     end
 
     -- Measurement class
-    if M.has_class(classes, "measurement") then
+    if has_class(classes, "measurement") then
         local text = pandoc.utils.stringify(inline.content)
         local value = tonumber(attributes.value) or tonumber(text:match("([%d%.]+)"))
         local unit = attributes.unit or text:match("%d+%.?%d*%s*(%a+)")
+        if not value then
+            io.stderr:write("Warning: measurement missing parseable value: " .. text .. "\n")
+            value = 0
+        end
+        if not unit then
+            io.stderr:write("Warning: measurement missing parseable unit: " .. text .. "\n")
+            unit = ""
+        end
         return {{
             type = "measurement_sentinel",
             value = value,
@@ -452,12 +455,6 @@ convert_inline = function(inline, marks, ctx)
     return {}
 end
 
--- Export convert_inline as module function
-M.convert_inline = convert_inline
-
--- Expose handlers table for testing
-M._handlers = inline_handlers
-
 -- Check if a node is a sentinel (non-text node that needs block-level handling)
 local function is_sentinel(node)
     return node.type and node.type:match("_sentinel$")

lib/metadata.lua

@@ -109,7 +109,8 @@ local function extract_authors(meta)
 
     local t = author.t or author.tag
 
-    if t == "MetaList" or (type(author) == "table" and #author > 0) then
+    -- MetaList is always a list of authors
+    if t == "MetaList" then
         local authors = {}
         for _, a in ipairs(author) do
             local author_obj = extract_single_author(a)
@@ -120,11 +121,29 @@ local function extract_authors(meta)
         if #authors > 0 then
             return authors
         end
-    else
-        -- Single author
-        local author_obj = extract_single_author(author)
-        if author_obj then
-            return { author_obj }
+        return nil
+    end
+
+    -- MetaInlines, MetaString, or Pandoc 3.x Inlines → single author
+    -- Try extracting as a single author first; this handles both tagged
+    -- MetaInlines/MetaString and Pandoc 3.x Inlines (which have no tag
+    -- but are stringify-able)
+    local single = extract_single_author(author)
+    if single then
+        return { single }
+    end
+
+    -- Fallback: untagged table that isn't stringify-able → try as list
+    if type(author) == "table" and not t and #author > 0 then
+        local authors = {}
+        for _, a in ipairs(author) do
+            local author_obj = extract_single_author(a)
+            if author_obj then
+                table.insert(authors, author_obj)
+            end
+        end
+        if #authors > 0 then
+            return authors
         end
     end
 

lib/reader_blocks.lua

@@ -188,7 +188,8 @@ function M.convert_list_item(item)
     for _, child in ipairs(item.children or {}) do
         local converted = M.convert_block(child)
         if converted then
-            if type(converted) == "table" and converted.tag then
+            -- Pandoc elements are userdata with t/tag accessors, not plain tables
+            if converted.t or converted.tag then
                 table.insert(blocks, converted)
             elseif type(converted) == "table" and #converted > 0 then
                 for _, b in ipairs(converted) do