Rework tree generation mode and API

Store tree nodes in congruent, compact arena array.
Define handle types to refer to tree nodes rather than pointers to
structure instances.
Free tree with context.

assets/parser.d.erb

@@ -105,22 +105,23 @@ public <%= typestring %> <%= @grammar.prefix %>value_get<%= name == "default" ?
 <% end %>
 
 <% if @grammar.tree %>
-/** Common tree node structure. */
-private struct TreeNode
-{
-    <%= @grammar.prefix %>position_t position;
-    <%= @grammar.prefix %>position_t end_position;
-    ushort n_fields;
-    bool is_token;
-    void *[0] fields;
-}
+/** Tree node ID type (index into the context node arena). ID 0 is null. */
+public alias <%= @grammar.prefix %>node_id_t = uint;
 
-/** Tree node types. @{ */
-public struct <%= @grammar.tree_prefix %>Token<%= @grammar.tree_suffix %>
+/**
+ * Tree node record.
+ *
+ * All tree nodes are stored contiguously in the context node arena. Child
+ * links are stored in a shared children array: a node's children
+ * occupy children[child_offset .. child_offset + n_fields]. Token payload
+ * fields (token, pvalue, and any user fields) are only meaningful when
+ * is_token is true.
+ */
+private struct <%= @grammar.prefix %>node_data_t
 {
-    /* TreeNode fields must be present in the same order here. */
     <%= @grammar.prefix %>position_t position;
     <%= @grammar.prefix %>position_t end_position;
+    <%= @grammar.prefix %>node_id_t child_offset;
     ushort n_fields;
     bool is_token;
     <%= @grammar.prefix %>token_t token;
@@ -128,22 +129,84 @@ public struct <%= @grammar.tree_prefix %>Token<%= @grammar.tree_suffix %>
 <%= @grammar.token_user_fields %>
 }
 
-<%   @parser.rule_sets.each do |name, rule_set| %>
-<%     next if name.start_with?("$") %>
-<%     next if rule_set.optional? %>
-public struct <%= @grammar.tree_prefix %><%= name %><%= @grammar.tree_suffix %>
+/** Tree node handle types. @{ */
+
+/** Token tree node handle. */
+public struct <%= @grammar.tree_prefix %>Token<%= @grammar.tree_suffix %>
 {
-    <%= @grammar.prefix %>position_t position;
-    <%= @grammar.prefix %>position_t end_position;
-    ushort n_fields;
-    bool is_token;
-<%     rule_set.tree_fields.each do |fields| %>
-    union
+    private <%= @grammar.prefix %>context_t * __context;
+    private <%= @grammar.prefix %>node_id_t __id;
+
+    this(<%= @grammar.prefix %>context_t * context, <%= @grammar.prefix %>node_id_t id)
     {
-<%       fields.each do |field_name, type| %>
-        <%= type %> * <%= field_name %>;
-<%       end %>
+        this.__context = context;
+        this.__id = id;
     }
+
+    /** Return whether this handle refers to a valid (non-null) node. */
+    @property bool valid()
+    {
+        return __id != 0u;
+    }
+
+    /** Access the underlying node record (token, pvalue, and user fields). */
+    @property ref <%= @grammar.prefix %>node_data_t __node()
+    {
+        return __context.<%= @grammar.prefix %>tree_nodes[__id];
+    }
+
+    alias __node this;
+}
+
+<%   tree_node_rule_sets.each do |rule_set| %>
+/** <%= rule_set.name %> tree node handle. */
+public struct <%= @grammar.tree_prefix %><%= rule_set.name %><%= @grammar.tree_suffix %>
+{
+    private <%= @grammar.prefix %>context_t * __context;
+    private <%= @grammar.prefix %>node_id_t __id;
+
+    this(<%= @grammar.prefix %>context_t * context, <%= @grammar.prefix %>node_id_t id)
+    {
+        this.__context = context;
+        this.__id = id;
+    }
+
+    /** Return whether this handle refers to a valid (non-null) node. */
+    @property bool valid()
+    {
+        return __id != 0u;
+    }
+
+    /** Text position of the first code point spanned by this node. */
+    @property <%= @grammar.prefix %>position_t position()
+    {
+        return __context.<%= @grammar.prefix %>tree_nodes[__id].position;
+    }
+
+    /** Text position of the last code point spanned by this node. */
+    @property <%= @grammar.prefix %>position_t end_position()
+    {
+        return __context.<%= @grammar.prefix %>tree_nodes[__id].end_position;
+    }
+
+    /** Number of child fields in this node. */
+    @property ushort n_fields()
+    {
+        return __id ? __context.<%= @grammar.prefix %>tree_nodes[__id].n_fields : cast(ushort)0u;
+    }
+<%     rule_set.tree_fields.each_with_index do |fields, i| %>
+<%       fields.each do |field_name, type| %>
+
+    /** Access the <%= field_name %> child node. */
+    @property <%= type %> <%= field_name %>()
+    {
+        if (__id == 0u)
+        {
+            return <%= type %>(__context, 0u);
+        }
+        return <%= type %>(__context, __context.<%= @grammar.prefix %>tree_children[__context.<%= @grammar.prefix %>tree_nodes[__id].child_offset + <%= i %>u]);
+    }
+<%       end %>
 <%     end %>
 }
 
@@ -196,7 +259,13 @@ public struct <%= @grammar.prefix %>context_t
 
     /** Parse result value. */
 <% if @grammar.tree %>
-    void * parse_result;
+    <%= @grammar.prefix %>node_id_t parse_result;
+
+    /** Tree node arena. Node ID 0 is reserved as the null node. */
+    <%= @grammar.prefix %>node_data_t[] <%= @grammar.prefix %>tree_nodes;
+
+    /** Shared tree child links. */
+    <%= @grammar.prefix %>node_id_t[] <%= @grammar.prefix %>tree_children;
 <% else %>
     <%= @grammar.prefix %>value_t parse_result;
 <% end %>
@@ -268,6 +337,11 @@ private enum size_t INVALID_ID = cast(size_t)-1;
     context.text_position.row = 1u;
     context.text_position.col = 1u;
     context.mode = <%= @lexer.mode_id("default") %>;
+<% if @grammar.tree %>
+
+    /* Reserve node ID 0 as the null tree node. */
+    context.<%= @grammar.prefix %>tree_nodes = new <%= @grammar.prefix %>node_data_t[](1);
+<% end %>
 
     return context;
 }
@@ -280,6 +354,16 @@ private enum size_t INVALID_ID = cast(size_t)-1;
  */
 void <%= @grammar.prefix %>context_delete(<%= @grammar.prefix %>context_t * context)
 {
+<% if @grammar.tree && @grammar.free_token_node != "" %>
+    foreach (ref node; context.<%= @grammar.prefix %>tree_nodes)
+    {
+        if (node.is_token)
+        {
+            <%= @grammar.prefix %>node_data_t * token_tree_node = &node;
+<%= expand_code(@grammar.free_token_node, false, nil, nil) %>
+        }
+    }
+<% end %>
 }
 
 /**************************************************************************
@@ -888,8 +972,8 @@ private struct state_value_t
     size_t state_id;
 
 <% if @grammar.tree %>
-    /** Tree node. */
-    void * tree_node;
+    /** Tree node ID. */
+    <%= @grammar.prefix %>node_id_t node_id;
 <% else %>
     <%= @grammar.prefix %>position_t position;
     <%= @grammar.prefix %>position_t end_position;
@@ -1007,7 +1091,7 @@ private <%= @grammar.prefix %>position_t get_rule_position(state_value_t[] state
  * @retval P_USER_TERMINATED
  *   User requested to terminate parsing.
  */
-private size_t parser_user_code(<%= @grammar.tree ? "void" : "#{@grammar.prefix}value_t" %> * _pvalue, uint rule, state_value_t[] statevalues, uint n_states, <%= @grammar.prefix %>context_t * context)
+private size_t parser_user_code(<%= @grammar.tree ? "#{@grammar.prefix}node_id_t _node_id" : "#{@grammar.prefix}value_t * _pvalue" %>, uint rule, state_value_t[] statevalues, uint n_states, <%= @grammar.prefix %>context_t * context)
 {
     switch (rule)
     {
@@ -1112,7 +1196,7 @@ private size_t parse_from(<%= @grammar.prefix %>context_t * context, size_t star
     size_t reduced_rule_set = INVALID_ID;
     size_t last_shifted_rule_set_id = INVALID_ID;
 <% if @grammar.tree %>
-    void * reduced_parser_node;
+    <%= @grammar.prefix %>node_id_t reduced_parser_node;
 <% else %>
     <%= @grammar.prefix %>position_t reduced_position;
     <%= @grammar.prefix %>position_t reduced_end_position;
@@ -1153,7 +1237,7 @@ private size_t parse_from(<%= @grammar.prefix %>context_t * context, size_t star
             {
                 /* Successful parse. */
 <% if @grammar.tree %>
-                context.parse_result = statevalues[$-1].tree_node;
+                context.parse_result = statevalues[$-1].node_id;
 <% else %>
                 context.parse_result = statevalues[$-1].pvalue;
 <% end %>
@@ -1179,7 +1263,7 @@ private size_t parse_from(<%= @grammar.prefix %>context_t * context, size_t star
                     context.input_index -= token_info.length;
                     context.text_position = token_info.position;
 <% if @grammar.tree %>
-                    context.parse_result = statevalues[$-1].tree_node;
+                    context.parse_result = statevalues[$-1].node_id;
 <% else %>
                     context.parse_result = statevalues[$-1].pvalue;
 <% end %>
@@ -1199,9 +1283,17 @@ private size_t parse_from(<%= @grammar.prefix %>context_t * context, size_t star
             {
                 /* We shifted a token, mark it consumed. */
 <% if @grammar.tree %>
-                <%= @grammar.tree_prefix %>Token<%= @grammar.tree_suffix %> * token_tree_node = new <%= @grammar.tree_prefix %>Token<%= @grammar.tree_suffix %>(token_info.position, token_info.end_position, 0u, true, token, token_info.pvalue);
+                <%= @grammar.prefix %>node_id_t token_node_id = cast(<%= @grammar.prefix %>node_id_t)context.<%= @grammar.prefix %>tree_nodes.length;
+                context.<%= @grammar.prefix %>tree_nodes ~= <%= @grammar.prefix %>node_data_t.init;
+                <%= @grammar.prefix %>node_data_t * token_tree_node = &context.<%= @grammar.prefix %>tree_nodes[token_node_id];
+                token_tree_node.position = token_info.position;
+                token_tree_node.end_position = token_info.end_position;
+                token_tree_node.n_fields = 0u;
+                token_tree_node.is_token = true;
+                token_tree_node.token = token;
+                token_tree_node.pvalue = token_info.pvalue;
 <%= expand_code(@grammar.on_token_node, false, nil, nil) %>
-                statevalues[$-1].tree_node = token_tree_node;
+                statevalues[$-1].node_id = token_node_id;
 <% else %>
                 statevalues[$-1].position = token_info.position;
                 statevalues[$-1].end_position = token_info.end_position;
@@ -1213,7 +1305,7 @@ private size_t parse_from(<%= @grammar.prefix %>context_t * context, size_t star
             {
                 /* We shifted a RuleSet. */
 <% if @grammar.tree %>
-                statevalues[$-1].tree_node = reduced_parser_node;
+                statevalues[$-1].node_id = reduced_parser_node;
 <% else %>
                 statevalues[$-1].pvalue = reduced_parser_value;
                 statevalues[$-1].position = reduced_position;
@@ -1242,55 +1334,56 @@ private size_t parse_from(<%= @grammar.prefix %>context_t * context, size_t star
 <% if @grammar.tree %>
             if (parser_reduce_table[reduce_index].propagate_optional_target)
             {
-                reduced_parser_node = statevalues[$ - 1].tree_node;
+                reduced_parser_node = statevalues[$ - 1].node_id;
             }
             else if (parser_reduce_table[reduce_index].n_states > 0)
             {
-                size_t n_fields = parser_reduce_table[reduce_index].rule_set_node_field_array_size;
-                size_t node_size = TreeNode.sizeof + n_fields * (void *).sizeof;
-                TreeNode * node = cast(TreeNode *)malloc(node_size);
-                GC.addRange(node, node_size);
-                node.position = <%= @grammar.prefix %>position_t.INVALID;
-                node.end_position = <%= @grammar.prefix %>position_t.INVALID;
-                node.n_fields = cast(ushort)n_fields;
-                node.is_token = false;
-                foreach (i; 0..n_fields)
-                {
-                    node.fields[i] = null;
-                }
+                ushort n_fields = parser_reduce_table[reduce_index].rule_set_node_field_array_size;
+                /* Reserve child slots. New slots are zero-initialized
+                 * (null node ID) so absent optional children remain null. */
+                <%= @grammar.prefix %>node_id_t child_offset = cast(<%= @grammar.prefix %>node_id_t)context.<%= @grammar.prefix %>tree_children.length;
+                context.<%= @grammar.prefix %>tree_children.length += n_fields;
                 if (parser_reduce_table[reduce_index].rule_set_node_field_index_map is null)
                 {
                     foreach (i; 0..parser_reduce_table[reduce_index].n_states)
                     {
-                        node.fields[i] = statevalues[$ - parser_reduce_table[reduce_index].n_states + i].tree_node;
+                        context.<%= @grammar.prefix %>tree_children[child_offset + i] = statevalues[$ - parser_reduce_table[reduce_index].n_states + i].node_id;
                     }
                 }
                 else
                 {
                     foreach (i; 0..parser_reduce_table[reduce_index].n_states)
                     {
-                        node.fields[parser_reduce_table[reduce_index].rule_set_node_field_index_map[i]] = statevalues[$ - parser_reduce_table[reduce_index].n_states + i].tree_node;
+                        context.<%= @grammar.prefix %>tree_children[child_offset + parser_reduce_table[reduce_index].rule_set_node_field_index_map[i]] = statevalues[$ - parser_reduce_table[reduce_index].n_states + i].node_id;
                     }
                 }
+                <%= @grammar.prefix %>node_id_t node_id = cast(<%= @grammar.prefix %>node_id_t)context.<%= @grammar.prefix %>tree_nodes.length;
+                context.<%= @grammar.prefix %>tree_nodes ~= <%= @grammar.prefix %>node_data_t.init;
+                <%= @grammar.prefix %>node_data_t * node = &context.<%= @grammar.prefix %>tree_nodes[node_id];
+                node.position = <%= @grammar.prefix %>position_t.INVALID;
+                node.end_position = <%= @grammar.prefix %>position_t.INVALID;
+                node.child_offset = child_offset;
+                node.n_fields = n_fields;
+                node.is_token = false;
                 bool position_found = false;
                 foreach (i; 0..n_fields)
                 {
-                    TreeNode * child = cast(TreeNode *)node.fields[i];
-                    if (child && child.position.valid)
+                    <%= @grammar.prefix %>node_id_t child_id = context.<%= @grammar.prefix %>tree_children[child_offset + i];
+                    if (child_id != 0u && context.<%= @grammar.prefix %>tree_nodes[child_id].position.valid)
                     {
                         if (!position_found)
                         {
-                            node.position = child.position;
+                            node.position = context.<%= @grammar.prefix %>tree_nodes[child_id].position;
                             position_found = true;
                         }
-                        node.end_position = child.end_position;
+                        node.end_position = context.<%= @grammar.prefix %>tree_nodes[child_id].end_position;
                     }
                 }
-                reduced_parser_node = node;
+                reduced_parser_node = node_id;
             }
             else
             {
-                reduced_parser_node = null;
+                reduced_parser_node = 0u;
             }
 <%   if @grammar.parser_user_code_used? %>
             if (parser_user_code(reduced_parser_node, parser_reduce_table[reduce_index].rule, statevalues, parser_reduce_table[reduce_index].n_states, context) == P_USER_TERMINATED)
@@ -1360,14 +1453,14 @@ public size_t <%= @grammar.prefix %>parse_inner_<%= start_rule %>(<%= @grammar.p
  * @return Parse result value.
  */
 <% if @grammar.tree %>
-public <%= @grammar.tree_prefix %><%= @grammar.start_rules[0] %><%= @grammar.tree_suffix %> * <%= @grammar.prefix %>result(<%= @grammar.prefix %>context_t * context)
+public <%= @grammar.tree_prefix %><%= @grammar.start_rules[0] %><%= @grammar.tree_suffix %> <%= @grammar.prefix %>result(<%= @grammar.prefix %>context_t * context)
 {
-    return cast(<%= @grammar.tree_prefix %><%= @grammar.start_rules[0] %><%= @grammar.tree_suffix %> *)context.parse_result;
+    return <%= @grammar.tree_prefix %><%= @grammar.start_rules[0] %><%= @grammar.tree_suffix %>(context, context.parse_result);
 }
 <%   @grammar.start_rules.each_with_index do |start_rule, i| %>
-public <%= @grammar.tree_prefix %><%= start_rule %><%= @grammar.tree_suffix %> * <%= @grammar.prefix %>result_<%= start_rule %>(<%= @grammar.prefix %>context_t * context)
+public <%= @grammar.tree_prefix %><%= start_rule %><%= @grammar.tree_suffix %> <%= @grammar.prefix %>result_<%= start_rule %>(<%= @grammar.prefix %>context_t * context)
 {
-    return cast(<%= @grammar.tree_prefix %><%= start_rule %><%= @grammar.tree_suffix %> *)context.parse_result;
+    return <%= @grammar.tree_prefix %><%= start_rule %><%= @grammar.tree_suffix %>(context, context.parse_result);
 }
 <%   end %>
 <% else %>
@@ -1383,34 +1476,6 @@ public <%= start_rule_type(i)[1] %> <%= @grammar.prefix %>result_<%= start_rule
 <%   end %>
 <% end %>
 
-<% if @grammar.tree %>
-private void tree_delete(TreeNode * node)
-{
-    if (!node.is_token)
-    {
-        for (size_t i = 0u; i < node.n_fields; i++)
-        {
-            if (node.fields[i])
-            {
-                tree_delete(cast(TreeNode *)node.fields[i]);
-            }
-        }
-        GC.removeRange(node);
-        free(node);
-    }
-}
-
-void <%= @grammar.prefix %>tree_delete(<%= @grammar.tree_prefix %><%= @grammar.start_rules[0] %><%= @grammar.tree_suffix %> * tree)
-{
-    tree_delete(cast(TreeNode *)tree);
-}
-<%   @grammar.start_rules.each_with_index do |start_rule, i| %>
-void <%= @grammar.prefix %>tree_delete_<%= start_rule %>(<%= @grammar.tree_prefix %><%= start_rule %><%= @grammar.tree_suffix %> * tree)
-{
-    tree_delete(cast(TreeNode *)tree);
-}
-<%   end %>
-<% end %>
 
 /**
  * Get the current text input position.