markbridge 0.1.1 → 0.1.3

data/lib/markbridge/processors/discourse_markdown/detectors/upload.rb

@@ -22,11 +22,18 @@ module Markbridge
         #   match = detector.detect(input, 0)
         #   match.node.type # => :attachment
         class Upload < Base
-          # Pattern for image: ![alt|dimensions](upload://sha1.ext)
-          IMAGE_PATTERN = %r{!\[([^\]]*)\]\(upload://([^)]+)\)}
-
-          # Pattern for attachment: [filename|attachment](upload://sha1.ext) followed by optional (size)
-          ATTACHMENT_PATTERN = %r{\[([^\]]*\|attachment)\]\(upload://([^)]+)\)(\s*\([^)]+\))?}
+          # Image: ![alt|dimensions](upload://sha1.ext)
+          IMAGE_PATTERN =
+            %r{\A!\[(?<alt>[^|\]]*)(?:\|(?<dimensions>[^\]]*))?\]\(upload://(?<url>[^)]+)\)}
+
+          # Attachment: [filename|attachment](upload://sha1.ext) (size)
+          ATTACHMENT_PATTERN =
+            %r{
+            \A
+            \[(?<filename>[^|\]]*)\|attachment\]
+            \(upload://(?<url>[^)]+)\)
+            (?:\s*\((?<size>[^)]+)\))?
+          }xi
 
           # Attempt to detect an upload at the given position.
           #
@@ -34,14 +41,11 @@ module Markbridge
           # @param pos [Integer] current position to check
           # @return [Match, nil] match result or nil if no match
           def detect(input, pos)
-            char = input[pos]
-            return nil unless char == "!" || char == "["
-
             remaining = input[pos..]
-
-            if char == "!"
+            case input[pos]
+            when "!"
               detect_image(remaining, pos)
-            else
+            when "["
               detect_attachment(remaining, pos)
             end
           end
@@ -50,71 +54,42 @@ module Markbridge
 
           def detect_image(remaining, pos)
             match = IMAGE_PATTERN.match(remaining)
-            return nil unless match&.begin(0)&.zero?
-
-            raw = match[0]
-            alt_part = match[1]
-            url_part = match[2]
+            return nil unless match
 
-            # Parse alt and dimensions from "alt|dimensions" format
-            alt, dimensions = parse_alt_dimensions(alt_part)
+            sha1, filename = parse_upload_url(match[:url])
+            alt = match[:alt]
+            alt = nil if alt.empty?
 
-            # Extract SHA1 and filename from URL
-            sha1, filename = parse_upload_url(url_part)
+            # `type: :image` is omitted because it is AST::Upload's default -
+            # passing it explicitly was an equivalent-mutation surface.
+            node =
+              AST::Upload.new(sha1:, filename:, alt:, dimensions: match[:dimensions], raw: match[0])
 
-            node = AST::Upload.new(sha1:, filename:, type: :image, alt:, dimensions:, raw:)
-
-            Match.new(start_pos: pos, end_pos: pos + raw.length, node:)
+            Match.new(start_pos: pos, end_pos: pos + match[0].length, node:)
           end
 
           def detect_attachment(remaining, pos)
             match = ATTACHMENT_PATTERN.match(remaining)
-            return nil unless match&.begin(0)&.zero?
-
-            raw = match[0]
-            name_part = match[1]
-            url_part = match[2]
-            size_part = match[3]
-
-            # Parse filename from "filename|attachment" format
-            filename = name_part.sub(/\|attachment$/i, "")
+            return nil unless match
 
-            # Extract SHA1 from URL
-            sha1, _url_filename = parse_upload_url(url_part)
+            sha1, = parse_upload_url(match[:url])
 
-            # Parse size if present
-            size = size_part&.strip&.delete_prefix("(")&.delete_suffix(")")
+            node =
+              AST::Upload.new(
+                sha1:,
+                filename: match[:filename],
+                type: :attachment,
+                size: match[:size],
+                raw: match[0],
+              )
 
-            node = AST::Upload.new(sha1:, filename:, type: :attachment, size:, raw:)
-
-            Match.new(start_pos: pos, end_pos: pos + raw.length, node:)
-          end
-
-          def parse_alt_dimensions(alt_part)
-            return nil, nil if alt_part.nil? || alt_part.empty?
-
-            if alt_part.include?("|")
-              parts = alt_part.split("|", 2)
-              alt = parts[0].empty? ? nil : parts[0]
-              dimensions = parts[1]
-              [alt, dimensions]
-            else
-              [alt_part, nil]
-            end
+            Match.new(start_pos: pos, end_pos: pos + match[0].length, node:)
           end
 
+          # URL format: sha1.ext or just sha1. Returns [sha1, filename-or-nil].
           def parse_upload_url(url_part)
-            # URL format: sha1.ext or just sha1
-            if url_part.include?(".")
-              parts = url_part.split(".", 2)
-              sha1 = parts[0]
-              filename = url_part
-            else
-              sha1 = url_part
-              filename = nil
-            end
-
-            [sha1, filename]
+            sha1, _, ext = url_part.partition(".")
+            [sha1, ext.empty? ? nil : url_part]
           end
         end
       end

data/lib/markbridge/processors/discourse_markdown/scanner.rb

@@ -49,13 +49,8 @@ module Markbridge
         def initialize(detectors: DEFAULT_DETECTORS, tag_library: nil, mention_resolver: nil)
           @detector_instances = build_detectors(detectors, mention_resolver)
           @tag_library = tag_library
-          @code_tracker = nil
-          @result = nil
-          @nodes = nil
-          @node_index = 0
-          @pos = 0
-          @input = nil
-          @line_start = true
+          # @code_tracker / @result / @nodes / @node_index / @pos / @input /
+          # @line_start are set by #scan before use; no defensive init needed.
         end
 
         # Scan input and extract constructs.
@@ -63,14 +58,12 @@ module Markbridge
         # @param input [String] Discourse Markdown input
         # @return [ScanResult] result containing processed markdown and extracted nodes
         def scan(input)
-          return ScanResult.new(markdown: "", nodes: []) if input.nil? || input.empty?
-
           @code_tracker = CodeBlockTracker.new
           @result = +""
           @nodes = []
           @node_index = 0
           @pos = 0
-          @input = input
+          @input = input.to_s
           @line_start = true
 
           scan_input
@@ -82,14 +75,10 @@ module Markbridge
 
         def build_detectors(detectors, mention_resolver)
           detectors.map do |klass|
-            if klass.is_a?(Class)
-              if klass == Detectors::Mention && mention_resolver
-                klass.new(type_resolver: mention_resolver)
-              else
-                klass.new
-              end
+            if klass == Detectors::Mention
+              klass.new(type_resolver: mention_resolver)
             else
-              klass
+              klass.new
             end
           end
         end
@@ -102,9 +91,10 @@ module Markbridge
               next if advance_code_boundary(:check_indented_boundary)
             end
 
-            # Check for inline code boundary
-            if @input[@pos] == "`" && !@code_tracker.in_fenced_block &&
-                 !@code_tracker.in_indented_block
+            # Check for inline code boundary. check_inline_boundary's
+            # own fenced/indented guard means we don't need to pre-check
+            # here — it'll just return nil in those cases.
+            if @input[@pos] == "`"
               new_pos = @code_tracker.check_inline_boundary(@input, @pos)
               if new_pos
                 @result << @input[@pos...new_pos]
@@ -142,9 +132,15 @@ module Markbridge
           new_pos = @code_tracker.public_send(method, @input, @pos, line_start: true)
           return false unless new_pos
 
+          # check_fenced_boundary / check_indented_boundary always stop
+          # at pos_after_line, which is either after a "\n" or at EOF.
+          # After-newline → @line_start should be true; at EOF the
+          # outer `while @pos < @input.length` exits and @line_start
+          # is unobservable. Setting true unconditionally drops the
+          # `@input[new_pos - 1] == "\n"` dance.
           @result << @input[@pos...new_pos]
           @pos = new_pos
-          @line_start = new_pos > 0 && @input[new_pos - 1] == "\n"
+          @line_start = true
           true
         end
 
@@ -159,26 +155,24 @@ module Markbridge
         def handle_match(match)
           node = match.node
           @nodes << node
+          @result << render_placeholder(node)
 
-          # Render placeholder using tag library if available
-          placeholder = render_placeholder(node)
-          @result << placeholder
-
+          # Every detector shipped today matches content that ends on a
+          # non-newline byte (`]`, `)`, `_`, alphanumeric), so @line_start
+          # is always false after a successful match. If a future custom
+          # detector produces a match whose end_pos sits right after
+          # "\n", re-introduce the `@input[@pos - 1] == "\n"` check.
           @pos = match.end_pos
-          @line_start = @pos > 0 && @input[@pos - 1] == "\n"
+          @line_start = false
           @node_index += 1
         end
 
         def render_placeholder(node)
           if @tag_library
             tag = @tag_library[node.class]
-            if tag
-              # Create a minimal interface for rendering
-              return tag.render(node, nil)
-            end
+            return tag.render(node, nil) if tag
           end
 
-          # Default placeholder format if no tag library or tag not found
           default_placeholder(node)
         end
 
@@ -192,8 +186,6 @@ module Markbridge
             "<<EVENT:#{@node_index}:#{node.name}>>"
           when AST::Upload
             "<<UPLOAD:#{@node_index}:#{node.sha1}>>"
-          else
-            "<<UNKNOWN:#{@node_index}>>"
           end
         end
       end

data/lib/markbridge/renderers/discourse/builders/list_item_builder.rb

@@ -15,12 +15,10 @@ module Markbridge
           # @return [String]
           def build(content, marker:, indent:)
             lines = content.split("\n")
-            lines = [""] if lines.empty? # Handle empty content
             first_line = "#{indent}#{marker}#{lines.first}"
 
-            return "#{first_line}\n" if lines.size == 1
+            return "#{first_line}\n" if lines.size < 2
 
-            # Handle multi-line content with sophisticated blank line handling
             format_multiline(lines, first_line, indent)
           end
 
@@ -63,15 +61,17 @@ module Markbridge
             end
           end
 
-          # Handle empty lines in continuation
+          # Handle empty lines in continuation. Caller (format_continuation_line)
+          # only invokes this when `line.empty?`, and `content.split("\n")`
+          # trims trailing empty strings, so the LAST continuation line is
+          # never empty — `idx + 1` is always in bounds when we get here.
           # @param idx [Integer] index in continuation_lines
           # @param continuation_lines [Array<String>] all continuation lines
           # @param continuation_indent [String] indent for continuation
           # @return [String, nil] formatted line or nil to skip
           def handle_empty_line(idx, continuation_lines, continuation_indent)
             # Skip empty lines that come before nested list items (structural blanks)
-            next_line = continuation_lines[idx + 1]
-            return nil if next_line&.match?(/\A\s*(?:-|\d+\.)\s/)
+            return nil if continuation_lines[idx + 1].match?(/\A\s*(?:-|\d+\.)\s/)
 
             # Preserve empty lines within text content (paragraph breaks) with indentation
             continuation_indent

data/lib/markbridge/renderers/discourse/html_escaper.rb

@@ -0,0 +1,20 @@
+# frozen_string_literal: true
+
+require "cgi"
+
+module Markbridge
+  module Renderers
+    module Discourse
+      # Escapes text for safe inclusion in HTML output. Used when rendering
+      # content inside a CommonMark HTML block (e.g. TableTag's fallback)
+      # where Markdown-level escaping would not be applied.
+      class HtmlEscaper
+        # @param text [String, nil]
+        # @return [String]
+        def self.escape(text)
+          CGI.escapeHTML(text || "")
+        end
+      end
+    end
+  end
+end

data/lib/markbridge/renderers/discourse/markdown_escaper.rb

@@ -36,18 +36,15 @@ module Markbridge
         #   breaks disabled by default.
         def initialize(escape_hard_line_breaks: false)
           @escape_hard_line_breaks = escape_hard_line_breaks
-          @inline_content = nil
-          @inline_result = nil
-          @inline_len = 0
+          # @inline_content / @inline_result / @inline_len are set by
+          # escape_inline on every call before any helper reads them;
+          # no defensive init needed.
         end
 
-        # Fast-path check: any character that might need escaping
-        # Only includes characters we actually escape (removed ], {, }, ^)
-        # > is needed for blockquote detection at line start
+        # Fast-path: skip escape_text entirely for content with no special
+        # chars. `>` is needed for blockquote detection at line start.
         MAYBE_SPECIAL = /[\\`*_\[#+\-.!<>&|~=>)]/
 
-        # Check for indented code on any line
-        # Matches: 4+ spaces, tab, or space+tab combinations that reach column 4+
         MAYBE_INDENTED_CODE = /(?:^|\n)(?: {4}|\t| {1,3}\t)/
 
         # Block-level patterns
@@ -122,8 +119,7 @@ module Markbridge
         # @return [String] the escaped text, or empty string if input is nil
         # @note Multi-line HTML tags and blocks are handled by escaping the opening <
         def escape(text)
-          return "".freeze if text.nil?
-          return text if text.empty?
+          return "" if text.nil?
 
           # Neutralize hard line breaks (trailing 2+ spaces before newline)
           text = text.gsub(/  +\n/, "\n") if @escape_hard_line_breaks && text.include?("  \n")
@@ -136,7 +132,14 @@ module Markbridge
         private
 
         def escape_text(text)
-          lines = text.split("\n", -1)
+          # On CRLF input, consume `\r` as part of the line terminator instead
+          # of leaving it on the line. A trailing `\r` breaks line-end anchored
+          # regexes (e.g. SETEXT_UNDERLINE_*) and the `ws_end >= line_length`
+          # early-out in escape_indented_code, leaking NBSPs onto
+          # whitespace-only CRLF lines. The `include?` guard keeps the
+          # LF-only fast path on a string split (regex split is ~20% slower
+          # on the indented-code hot path).
+          lines = text.include?("\r") ? text.split(/\r?\n/, -1) : text.split("\n", -1)
           return escape_line(lines[0], false) if lines.size == 1
 
           # Pre-allocate result buffer
@@ -158,37 +161,32 @@ module Markbridge
         end
 
         def escape_line(line, prev_was_paragraph)
-          return line if line.empty?
-
-          # Handle indented code blocks first
+          # No `line.empty?` early-return: it's redundant with the
+          # `line.getbyte(indent_len).nil?` guard below, which catches both
+          # empty and whitespace-only lines while also preserving object
+          # identity (returns `line`).
           return escape_indented_code(line) if INDENTED_CODE.match?(line)
 
-          # Extract 0-3 space indent
-          line_length = line.length
+          # After INDENTED_CODE, line has at most 3 leading spaces, so the
+          # `< 3` bound keeps this a tight YJIT-friendly hot loop.
           indent_len = 0
-          while indent_len < 3 && indent_len < line_length && line.getbyte(indent_len) == SPACE
-            indent_len += 1
-          end
+          indent_len += 1 while indent_len < 3 && line.getbyte(indent_len) == SPACE
 
-          return line if indent_len >= line_length
+          # Whitespace-only line (1-3 spaces) — getbyte past end is nil.
+          return line if line.getbyte(indent_len).nil?
 
           has_indent = indent_len > 0
           content = has_indent ? line[indent_len..] : line
 
-          # Apply block-level escaping (which may also do inline escaping)
           escaped, skip_inline = escape_block_level(content, prev_was_paragraph)
-
-          # Apply inline escaping if block-level didn't handle it
           escaped = escape_inline(escaped) unless skip_inline
 
-          # Prepend indent if present, preserve encoding
           if has_indent
-            encoding = line.encoding
-            result = String.new(encoding:)
+            result = String.new(encoding: line.encoding)
             result << line[0, indent_len] << escaped
             result
           else
-            escaped.is_a?(String) ? escaped.force_encoding(line.encoding) : escaped
+            escaped.force_encoding(line.encoding)
           end
         end
 
@@ -203,15 +201,14 @@ module Markbridge
           # - Content doesn't start at valid block position (no lists, headings, etc.)
           # - Visual indentation is preserved (NBSP renders as space)
           # We still escape inline content since it's no longer protected.
+          # Caller (escape_line) guarantees INDENTED_CODE matched, so line
+          # starts with at least one SPACE or TAB; ws_end is always ≥ 1.
           line_length = line.length
           ws_end = 0
-          while ws_end < line_length
-            byte = line.getbyte(ws_end)
-            break if byte != SPACE && byte != TAB
+          while ws_end < line_length && ((byte = line.getbyte(ws_end)) == SPACE || byte == TAB)
             ws_end += 1
           end
 
-          return line if ws_end == 0 # No leading whitespace (shouldn't happen, but safe)
           return line if ws_end >= line_length # Whitespace-only line
 
           # Convert leading whitespace to NBSP (tab = 4 NBSP for visual consistency)
@@ -310,6 +307,13 @@ module Markbridge
           @inline_len = bytesize
           pos = 0
 
+          # No loop-progress guard: every `dispatch_inline_byte` branch
+          # returns `pos + N` for N >= 1 by construction, so the loop
+          # is provably terminating. Mutations that break this
+          # (`while true`, body drops, selector swaps that short-circuit
+          # the dispatch) surface as timeouts rather than alive
+          # mutations, and the inline guard would otherwise cost ~15%
+          # on this hot path per benchmark.
           while pos < @inline_len
             byte = @inline_content.getbyte(pos)
             pos = dispatch_inline_byte(byte, pos)
@@ -474,39 +478,42 @@ module Markbridge
         end
 
         def paragraph_line?(line)
-          return false if line.empty?
+          pos = 0
+          line_len = line.bytesize
+          pos += 1 while pos < line_len && line.getbyte(pos) == SPACE
+          first_non_space = pos
 
-          line_length = line.length
-          first_non_space = 0
-          while first_non_space < line_length && line.getbyte(first_non_space) == SPACE
-            first_non_space += 1
-          end
-          return false if first_non_space >= line_length || line.getbyte(first_non_space) == TAB
+          # Empty or whitespace-only lines: getbyte past the end returns nil.
+          return false if line.getbyte(first_non_space).nil?
 
-          content = first_non_space <= 3 ? line[first_non_space..] : line
+          # Indented code (4+ spaces or any leading \t) is not a paragraph.
+          # INDENTED_CODE also catches lines where first_non_space > 3, so no
+          # separate numeric boundary check is needed.
+          return false if INDENTED_CODE.match?(line)
 
-          # Lines starting with [ get escaped to \[, which IS paragraph content
-          # So setext headings CAN follow them
-          return true if content.getbyte(0) == BRACKET_OPEN
+          content = first_non_space == 0 ? line : line[first_non_space..]
 
-          !block_construct?(content) && !INDENTED_CODE.match?(line)
+          # Lines starting with [ are paragraph content (the escaper rewrites [
+          # to \[). block_construct? has no BRACKET_OPEN case arm, so such
+          # lines naturally fall through and !block_construct?(content) == true.
+          !block_construct?(content)
         end
 
         # Checks whether content starts with a block-level markdown construct.
         # Used by both escape_block_level (to decide what to escape) and
         # paragraph_line? (to decide if setext underlines can follow).
         def block_construct?(content)
-          first_byte = content.getbyte(0)
-
-          case first_byte
+          case content.getbyte(0)
           when HASH
             ATX_HEADING.match?(content)
           when GT
             true
-          when DASH, PLUS, STAR
-            BULLET_LIST.match?(content) ||
-              (first_byte == DASH && THEMATIC_BREAK_DASH.match?(content)) ||
-              (first_byte == STAR && THEMATIC_BREAK_STAR.match?(content))
+          when DASH
+            BULLET_LIST.match?(content) || THEMATIC_BREAK_DASH.match?(content)
+          when STAR
+            BULLET_LIST.match?(content) || THEMATIC_BREAK_STAR.match?(content)
+          when PLUS
+            BULLET_LIST.match?(content)
           when UNDERSCORE
             THEMATIC_BREAK_UNDERSCORE.match?(content)
           when BACKTICK

data/lib/markbridge/renderers/discourse/render_context.rb

@@ -11,26 +11,40 @@ module Markbridge
       class RenderContext
         attr_reader :parents, :depth
 
-        def initialize(parents = [], parent_cache: nil)
+        def initialize(parents = [], parent_cache: nil, html_mode: false)
           @parents = parents.freeze
           @depth = parents.size
           @parent_cache = parent_cache || build_cache(parents)
+          @html_mode = html_mode
         end
 
-        # Create new context with element added to parent chain
-        # Incrementally updates cache instead of rebuilding from scratch
+        # Create new context with element added to parent chain.
+        # Incrementally updates the cache (O(1)) instead of rebuilding from
+        # parents (O(depth)) — important for deeply-nested documents.
         # @param element [AST::Element]
         # @return [RenderContext]
         def with_parent(element)
           new_parents = @parents + [element]
 
-          # Incrementally update cache instead of rebuilding
           new_cache = @parent_cache.dup
           element_class = element.class
           new_cache[element_class] ||= []
           new_cache[element_class] = new_cache[element_class] + [element]
 
-          self.class.new(new_parents, parent_cache: new_cache)
+          self.class.new(new_parents, parent_cache: new_cache, html_mode: @html_mode)
+        end
+
+        # Create new context with html_mode toggled
+        # Preserves parent chain and cache
+        # @param value [Boolean]
+        # @return [RenderContext]
+        def with_html_mode(value)
+          self.class.new(@parents, parent_cache: @parent_cache, html_mode: value)
+        end
+
+        # @return [Boolean]
+        def html_mode?
+          @html_mode
         end
 
         # Find closest parent of given type
@@ -54,7 +68,7 @@ module Markbridge
         # @param klass [Class]
         # @return [Boolean]
         def has_parent?(klass)
-          @parent_cache.key?(klass) && !@parent_cache[klass].empty?
+          !@parent_cache[klass].nil?
         end
 
         # Check if we're at the root (no parents)
@@ -65,14 +79,12 @@ module Markbridge
 
         private
 
-        # Build cache from parents array
-        # Groups parents by class for fast lookup
+        # Build cache from parents array.
+        # Groups parents by class for fast O(1) lookup.
         # @param parents [Array<AST::Element>]
         # @return [Hash{Class => Array<AST::Element>}]
         def build_cache(parents)
-          parents.each_with_object(Hash.new { |h, k| h[k] = [] }) do |parent, cache|
-            cache[parent.class] = cache[parent.class] + [parent]
-          end
+          parents.group_by(&:class)
         end
       end
     end