@zalom/plastic 1.0.0-beta.8 → 1.0.0

package/scripts/lib/links_projection.rb

@@ -0,0 +1,160 @@
+# encoding: UTF-8
+# frozen_string_literal: true
+
+# LinksProjection — pure logic that projects one intent's sources/chain graph into
+# the canonical I5 `## Links` section text (intent 72). Mirrors the pure-module
+# style of GraphRebuild: module-function helpers, no file IO, no eval, no
+# ENV/global state. The IO shell (scripts/project-links) and doctor build the
+# cross-store resolver and feed it here.
+#
+# Pinned canonical projection (the human's format call):
+#   1. ORDERING (load-bearing): ALL sources first, in frontmatter order, THEN all
+#      chain, in frontmatter order. Sources can NEVER appear at the end. No group
+#      headings, no per-entry source/chain tags: the ordering carries the meaning.
+#   2. ENTRY SHAPE: one list item `- [[<id>--<slug>|<target's full intent: text>]]`,
+#      where the wikilink TARGET is the target intent's resolvable `id--slug` file
+#      basename (so it clicks through in Obsidian) and the LABEL is the target's
+#      full `intent:` frontmatter text, whitespace-trimmed.
+#   3. CROSS-STORE: a cross-store target renders
+#      `- [[<store>:<id>--<slug>|<target's full intent: text>]]`.
+#   4. RESOLVER MISS: a ref that resolves to no intent raises UnresolvedRef. No
+#      bare-id, slug-less, or guessed link is ever emitted.
+#   5. EMPTY-STATE: empty sources AND chain yields the heading plus a single
+#      explanatory comment.
+module LinksProjection
+  module_function
+
+  HEADING = "## Links"
+  EMPTY_COMMENT = "<!-- No sources or chain; this intent has no graph edges to project. -->"
+
+  # Raised when a sources/chain ref resolves to no intent. Carries the offending
+  # ref so the IO shell can report it per-intent and skip the write.
+  class UnresolvedRef < StandardError
+    attr_reader :ref
+
+    def initialize(ref)
+      @ref = ref
+      super("unresolved sources/chain ref: #{ref.inspect}")
+    end
+  end
+
+  # PURE. Build the canonical `## Links` section text for one intent.
+  #
+  #   sources — array of id / `store:id` ref strings, in frontmatter order
+  #   chain   — array of id / `store:id` ref strings, in frontmatter order
+  #   resolve — a callable (`->(ref) { ... }`) mapping ONE ref to a Hash like
+  #             { target: "<id>--<slug>",            label: "<full intent: text>" } or
+  #             { target: "<store>:<id>--<slug>",    label: "<full intent: text>" }.
+  #             Returning nil (or a Hash lacking :target) signals no target and
+  #             raises UnresolvedRef. Keeping resolution injected keeps this module
+  #             pure and hermetically testable with in-memory maps.
+  #
+  # Returns the full section text: the `## Links` heading line, one entry line per
+  # ref (sources first, then chain), and a single trailing newline. The empty case
+  # returns the heading + the empty-state comment + a single trailing newline.
+  def section(sources:, chain:, resolve:)
+    src = Array(sources).map(&:to_s)
+    chn = Array(chain).map(&:to_s)
+
+    # Resolve EVERY ref to its { target:, label: } first, then dedup by the RESOLVED
+    # target (not the raw ref string). This is load-bearing: the same intent may be
+    # referenced as a bare id in one group and as `store:id` in another (or via a
+    # relocation), which dedups identically only AFTER resolution. Sources win
+    # (formative edge), and frontmatter order is preserved within each group.
+    seen = {}
+    rendered = []
+    src.each { |ref| add_entry(ref, resolve, seen, rendered) }
+    chn.each { |ref| add_entry(ref, resolve, seen, rendered) }
+
+    return empty_section if rendered.empty?
+
+    (["#{HEADING}\n"] + rendered.map { |line| "#{line}\n" }).join
+  end
+
+  # PURE. The canonical empty-state section: heading + the single comment line.
+  def empty_section
+    "#{HEADING}\n#{EMPTY_COMMENT}\n"
+  end
+
+  # Resolve `ref`, render its entry, and append it to `rendered` UNLESS its resolved
+  # target was already emitted (dedup by resolved target, first-seen wins so sources
+  # precede chain). Mutates `seen` and `rendered`. Raises UnresolvedRef on a miss.
+  def add_entry(ref, resolve, seen, rendered)
+    target, label = resolve_entry(ref, resolve)
+    return if seen.key?(target)
+
+    seen[target] = true
+    rendered << "- [[#{target}|#{label}]]"
+  end
+
+  # PURE. Resolve one ref to [target, label]. Raises UnresolvedRef when the
+  # resolver returns nothing usable.
+  def resolve_entry(ref, resolve)
+    resolved = resolve.call(ref)
+    target = resolved.is_a?(Hash) ? resolved[:target] || resolved["target"] : nil
+    raise UnresolvedRef, ref if target.nil? || target.to_s.strip.empty?
+
+    label = (resolved[:label] || resolved["label"]).to_s.strip
+    [target.to_s, label]
+  end
+
+  # PURE. Render one entry line `- [[<target>|<label>]]` from a single ref. Kept for
+  # callers/tests that render one entry; #section uses add_entry for dedup.
+  def entry(ref, resolve)
+    target, label = resolve_entry(ref, resolve)
+    "- [[#{target}|#{label}]]"
+  end
+
+  # PURE. Resolve ONE sources/chain ref to its `{ target:, label: }` projection,
+  # given the in-memory cross-store maps. This is the single resolver definition
+  # shared by the IO shell (scripts/project-links) and the doctor check, so the two
+  # can never diverge.
+  #
+  #   ref            — "40" (same-store bare id) or "knowdb:1" (cross-store)
+  #   referer_store  — store_key of the intent carrying the ref ("global" / "project:<slug>")
+  #   relocation_map — from GraphRebuild.build_relocation_map (spans all stores)
+  #   store_index    — { store_key => [bare ids present] } (spans all stores)
+  #   node_index     — { store_key => { id => { basename:, label: } } } (spans all stores)
+  #
+  # Returns { target:, label: } (target is `<id>--<slug>` for a same-store id, or
+  # `<slug>:<id>--<slug>` for a cross-store one), or nil when the ref resolves to no
+  # live intent (which makes #section / #entry raise UnresolvedRef).
+  #
+  # Uses GraphRebuild.resolve_ref so a relocation always wins over a coincidentally
+  # reused id (the `global:24` impostor hazard), exactly as the frontmatter rebuild
+  # and the cross-store doctor check do.
+  def resolve_ref_projection(ref, referer_store:, relocation_map:, store_index:, node_index:)
+    require_relative "graph_rebuild"
+
+    res = GraphRebuild.resolve_ref(ref, referer_store: referer_store,
+                                        relocation_map: relocation_map,
+                                        store_index: store_index)
+    case res[:status]
+    when :same_store
+      node = (node_index[referer_store] || {})[res[:id]]
+      return nil if node.nil?
+
+      { target: node[:basename], label: node[:label] }
+    when :cross_store
+      slug, bare = res[:ref].split(":", 2)
+      target_key = canonical_store_key(slug, store_index)
+      node = (node_index[target_key] || {})[bare]
+      return nil if node.nil?
+
+      { target: "#{slug}:#{node[:basename]}", label: node[:label] }
+    else # :dead
+      nil
+    end
+  end
+
+  # Map a ref store slug ("global", "knowdb", "plastic") to a node_index/store_index
+  # key ("global", "project:knowdb", "project:plastic"). Mirrors
+  # GraphRebuild.canonical_store_key's intent for the node_index keyspace.
+  def canonical_store_key(slug, store_index)
+    return "global" if slug == "global"
+    return slug if (store_index || {}).key?(slug)
+
+    projected = "project:#{slug}"
+    (store_index || {}).key?(projected) ? projected : slug
+  end
+end

package/scripts/lib/links_section.rb

@@ -0,0 +1,207 @@
+# encoding: UTF-8
+# frozen_string_literal: true
+
+require_relative "intent_validator"
+
+# LinksSection — pure, minimal, style-preserving rewrite of ONLY the `## Links`
+# section of an intent file's content string (intent 72). Mirrors the discipline of
+# FrontmatterWriter: pure (no file IO, no eval, no global/ENV state), and returns
+# the original content UNCHANGED when nothing changed (idempotency).
+#
+# `## Links` is canonically the LAST sanctioned section
+# (IntentValidator::SANCTIONED_SECTIONS = Intent, Context, Outcome, Insights,
+# Links), and real intents put it last. So a file with an existing `## Links` has
+# its section replaced in place (from the heading to the next top-level `## `
+# heading or EOF), and a file WITHOUT a `## Links` gets one appended at end-of-body,
+# separated by exactly one blank line, preserving the body's trailing-newline shape.
+#
+# FENCE AWARENESS (intent 72 corruption fix): a `## Links` heading INSIDE a fenced
+# code block (``` or ~~~, possibly with an info string like ```markdown) is part of
+# an EXAMPLE, not a real section. All section scanning here IGNORES headings inside
+# fences and only ever targets the REAL `## Links` section (outside any fence). The
+# section end is the next `## ` heading that is ALSO outside a fence, so a replace
+# never consumes or unbalances a code fence. If more than one REAL `## Links`
+# heading exists, #rewrite raises AmbiguousLinks rather than guess.
+#
+# Frontmatter is NEVER touched: the leading `---`...`---` block is preserved
+# byte-for-byte and only the body is rebuilt.
+module LinksSection
+  module_function
+
+  HEADING = "## Links"
+
+  # A fence delimiter: ``` or ~~~ (any length >= 3), optional leading whitespace,
+  # optional info string (e.g. ```markdown). Mirrors CommonMark fenced-code rules
+  # closely enough for intent bodies.
+  FENCE_RE = /\A\s*(`{3,}|~{3,})/
+
+  # Raised when a body has more than one REAL `## Links` heading outside any fence;
+  # the tool must fail loud rather than guess which one to rewrite.
+  class AmbiguousLinks < StandardError
+    def initialize(count)
+      super("found #{count} real `## Links` headings outside code fences; refusing to guess")
+    end
+  end
+
+  # PURE. Replace (or insert) the REAL `## Links` section in `content` with
+  # `section_text` (the canonical block from LinksProjection.section, which begins
+  # with the `## Links` heading line and ends with a single trailing newline).
+  # Returns the new content, or the original when nothing changed. Raises
+  # AmbiguousLinks when more than one real `## Links` heading exists.
+  def rewrite(content, section_text)
+    return content unless content.is_a?(String)
+
+    fm, body = split_frontmatter(content)
+    new_body = rewrite_body(body, section_text)
+    updated = "#{fm}#{new_body}"
+    updated == content ? content : updated
+  end
+
+  # Split content into [frontmatter_with_delimiters, body]. When there is no
+  # frontmatter block, the frontmatter part is "" and the whole content is the
+  # body. The frontmatter part is preserved byte-for-byte by the caller.
+  def split_frontmatter(content)
+    return ["", content] unless content.start_with?("---")
+
+    parts = content.split("---", 3)
+    return ["", content] if parts.length < 3
+
+    ["---#{parts[1]}---", parts[2]]
+  end
+
+  # Rewrite ONLY the REAL `## Links` section within the body text.
+  def rewrite_body(body, section_text)
+    bounds = links_bounds(body)
+    if bounds
+      replace_section(body, section_text, bounds)
+    else
+      insert_section(body, section_text)
+    end
+  end
+
+  # PURE. Locate the REAL `## Links` section (fence-aware). Returns
+  # [start_index, end_index] line indices into body.lines, where start_index is the
+  # `## Links` heading line and end_index is the index of the next out-of-fence
+  # `## ` heading (or lines.length at EOF). Returns nil when there is no real
+  # `## Links` heading. Raises AmbiguousLinks when more than one exists.
+  def links_bounds(body)
+    lines = body.to_s.lines
+    starts = real_links_heading_indices(lines)
+    return nil if starts.empty?
+    raise AmbiguousLinks, starts.length if starts.length > 1
+
+    start = starts.first
+    stop = next_out_of_fence_heading(lines, start + 1)
+    [start, stop]
+  end
+
+  # PURE. Indices of every `## Links` heading line that is OUTSIDE any code fence.
+  # Accepts a body String or an Array of lines.
+  def real_links_heading_indices(body_or_lines)
+    lines = body_or_lines.is_a?(Array) ? body_or_lines : body_or_lines.to_s.lines
+    indices = []
+    in_fence = false
+    fence_marker = nil
+    lines.each_with_index do |line, i|
+      if (m = fence_open_close(line, in_fence, fence_marker))
+        in_fence = m[:in_fence]
+        fence_marker = m[:marker]
+        next
+      end
+      indices << i if !in_fence && line.rstrip == HEADING
+    end
+    indices
+  end
+
+  # PURE. Index of the first `## ` heading at or after `from` that is OUTSIDE any
+  # code fence. Returns lines.length when none (EOF). Fence state is recomputed
+  # from the top so nested example fences after the real heading are respected.
+  def next_out_of_fence_heading(lines, from)
+    in_fence = false
+    fence_marker = nil
+    lines.each_with_index do |line, i|
+      if (m = fence_open_close(line, in_fence, fence_marker))
+        in_fence = m[:in_fence]
+        fence_marker = m[:marker]
+        next
+      end
+      return i if i >= from && !in_fence && line.start_with?("## ")
+    end
+    lines.length
+  end
+
+  # PURE. Given the current fence state, decide whether `line` is a fence delimiter
+  # and return the new state, or nil when the line is not a fence delimiter.
+  # An opening fence records its marker family (` or ~); a closing fence must use a
+  # marker of the SAME family and carry no info string.
+  def fence_open_close(line, in_fence, fence_marker)
+    m = line.match(FENCE_RE)
+    return nil unless m
+
+    marker = m[1]
+    family = marker[0] # "`" or "~"
+    if in_fence
+      # A closing fence uses the same family, length >= the opener, no info string.
+      rest = line.sub(FENCE_RE, "").strip
+      if family == fence_marker && rest.empty?
+        { in_fence: false, marker: nil }
+      else
+        # A delimiter of the OTHER family (or an info-string line) inside a fence is
+        # literal content, not a fence event.
+        nil
+      end
+    else
+      { in_fence: true, marker: family }
+    end
+  end
+
+  # True iff the body has a REAL (out-of-fence) `## Links` heading. Used by callers
+  # to classify regenerate-vs-add without re-deriving fence state.
+  def links_heading?(body)
+    !real_links_heading_indices(body.to_s.lines).empty?
+  end
+
+  # Replace the REAL `## Links` section (the [start, stop] line bounds) with
+  # `section_text`, preserving everything before the heading and after the section
+  # byte-for-byte (including any fenced example that lives BEFORE the real section).
+  def replace_section(body, section_text, bounds)
+    lines = body.lines
+    start, stop = bounds
+    before = lines[0...start].join
+    tail = (lines[stop..] || [])
+
+    # `section_text` already ends with exactly one newline. When there is trailing
+    # content (another section follows), separate the block from it with one blank
+    # line; otherwise the section ends the body.
+    block = tail.empty? ? section_text : "#{section_text}\n"
+    "#{before}#{block}#{tail.join}"
+  end
+
+  # Append a `## Links` section at end-of-body, after the last existing section,
+  # separated by exactly one blank line, preserving the body's trailing newline.
+  def insert_section(body, section_text)
+    trimmed = body.to_s.sub(/\s+\z/, "")
+    if trimmed.empty?
+      # An empty body (no sections) just becomes the section.
+      section_text
+    else
+      "#{trimmed}\n\n#{section_text}"
+    end
+  end
+
+  # PURE. Extract the REAL `## Links` section text (fence-aware), normalized to the
+  # canonical block shape the projection emits: heading line + entry lines + a
+  # single trailing newline. Returns "" when there is no real section. Shared by
+  # the IO shell's audit and the doctor drift check so all three agree on the
+  # location. Raises AmbiguousLinks when more than one real heading exists.
+  def extract_section(body)
+    bounds = links_bounds(body)
+    return "" if bounds.nil?
+
+    start, stop = bounds
+    lines = body.to_s.lines
+    section = lines[(start + 1)...stop].join.sub(/\n+\z/, "\n")
+    section = "" if section.strip.empty?
+    "#{HEADING}\n#{section}"
+  end
+end