[aux,init] move lazy char-def loading code into earlier stage of initialization

The fontloader requires parts of the ``characters`` table to be present
at load time. This turns out to interfere with our custom of installing
the lazy loader for the table components only after the fontloader has
been injected. Since inserting the code at the appropriate place in the
loading chain would be tedious and unmaintainable due to the various
load options, we just preinstall the metatable onto an empty table prior
to loading the loader.

Some precautions had to be taken regarding the ``classifiers`` subhash
of the table that needs to be relocated from the data we received via
mkcharacters.

2 files changed, 131 insertions, 129 deletions

diff --git a/src/luaotfload-auxiliary.lua b/src/luaotfload-auxiliary.lua
index 2c2624e..6ba21e0 100644
--- a/src/luaotfload-auxiliary.lua
+++ b/src/luaotfload-auxiliary.lua
@@ -298,132 +298,6 @@ end
 
 aux.name_of_slot      = name_of_slot
 
---[[doc--
-
-  In Context, characters.data is where the data from char-def.lua
-  resides. The file is huge (>4.4 MB as of 2016) and only a stripped
-  down version is part of the isolated font loader. Nevertheless, we
-  include an excerpt generated by the mkcharacters script that contains
-  a subset of the fields of each character defined and some extra
-  metadata.
-
-  Currently, these are (compare the mkcharacters script!)
-
-    · "direction"
-    · "mirror"
-    · "category"
-    · "textclass"
-
-  The directional information is required for packages like Simurgh [0]
-  to work correctly. In an early stage [1] it was necessary to load
-  further files from Context directly, including the full blown version
-  of char-def.  Since we have no use for most of the so imported
-  functionality, the required parts have been isolated and are now
-  instated along with luaotfload-characters.lua. We can extend the set
-  of imported features easily should it not be enough.
-
-  [0] https://github.com/persian-tex/simurgh
-  [1] http://tex.stackexchange.com/a/132301/14066
-
---doc]]--
-
-characters         = characters or { } --- should be created in basics-gen
-characters.data    = nil
-local chardef      = "luaotfload-characters"
-
-do
-  local setmetatableindex = function (t, f)
-    local mt = getmetatable (t)
-    if mt then
-      mt.__index = f
-    else
-      setmetatable (t, { __index = f })
-    end
-  end
-
-  --- there are some special tables for each field that provide access
-  --- to fields of the character table by means of a metatable
-
-  local mkcharspecial = function (characters, tablename, field)
-
-    local chardata = characters.data
-
-    if chardata then
-      local newspecial        = { }
-      characters [tablename]  = newspecial --> e.g. “characters.data.mirrors”
-
-      local idx = function (t, char)
-        local c = chardata [char]
-        if c then
-          local m = c [field] --> e.g. “mirror”
-          if m then
-            t [char] = m
-            return m
-          end
-        end
-        newspecial [char] = false
-        return char
-      end
-
-      setmetatableindex (newspecial, idx)
-    end
-
-  end
-
-  local mkcategories = function (characters) -- different from the others
-
-    local chardata = characters.data
-
-    setmetatable (characters, { __index = function (t, char)
-      if char then
-        local c = chardata [char]
-        c = c.category or char
-        t [char] = c
-        return c
-      end
-    end})
-
-  end
-
-  local load_failed = false
-  local chardata --> characters.data; loaded on demand
-
-  local load_chardef = function ()
-
-    logreport ("both", 1, "aux", "Loading character metadata from %s.", chardef)
-    chardata = dofile (kpse.find_file (chardef, "lua"))
-
-    if chardata == nil then
-      warning ("Could not load %s; continuing \z
-                with empty character table.",
-                chardef)
-      chardata    = { }
-      load_failed = true
-    end
-
-    characters      = { } --- nuke metatable
-    characters.data = chardata
-
-    --- institute some of the functionality from char-ini.lua
-
-    mkcharspecial (characters, "mirrors",     "mirror")
-    mkcharspecial (characters, "directions",  "direction")
-    mkcharspecial (characters, "textclasses", "textclass")
-    mkcategories  (characters)
-
-  end
-
-  local charindex = function (t, k)
-    if chardata == nil and load_failed ~= true then
-      load_chardef ()
-    end
-
-    return characters [k]
-  end
-
-  setmetatableindex (characters, charindex)
-
-end
 
 -----------------------------------------------------------------------
 ---                 features / scripts / languages
diff --git a/src/luaotfload-init.lua b/src/luaotfload-init.lua
index 096d778..5fafceb 100644
--- a/src/luaotfload-init.lua
+++ b/src/luaotfload-init.lua
@@ -319,6 +319,137 @@ local init_adapt = function ()
 
 end --- [init_adapt]
 
+--[[doc--
+
+  In Context, characters.data is where the data from char-def.lua
+  resides. The file is huge (>4.4 MB as of 2016) and only a stripped
+  down version is part of the isolated font loader. Nevertheless, we
+  include an excerpt generated by the mkcharacters script that contains
+  a subset of the fields of each character defined and some extra
+  metadata.
+
+  Currently, these are (compare the mkcharacters script!)
+
+    · "direction"
+    · "mirror"
+    · "category"
+    · "textclass"
+
+  The directional information is required for packages like Simurgh [0]
+  to work correctly. In an early stage [1] it was necessary to load
+  further files from Context directly, including the full blown version
+  of char-def.  Since we have no use for most of the so imported
+  functionality, the required parts have been isolated and are now
+  instated along with luaotfload-characters.lua. We can extend the set
+  of imported features easily should it not be enough.
+
+  [0] https://github.com/persian-tex/simurgh
+  [1] http://tex.stackexchange.com/a/132301/14066
+
+--doc]]--
+
+characters         = characters or { } --- should be created in basics-gen
+characters.data    = nil
+local chardef      = "luaotfload-characters"
+
+do
+  local setmetatableindex = function (t, f)
+    local mt = getmetatable (t)
+    if mt then
+      mt.__index = f
+    else
+      setmetatable (t, { __index = f })
+    end
+  end
+
+  --- there are some special tables for each field that provide access
+  --- to fields of the character table by means of a metatable
+
+  local mkcharspecial = function (characters, tablename, field)
+
+    local chardata = characters.data
+
+    if chardata then
+      local newspecial        = { }
+      characters [tablename]  = newspecial --> e.g. “characters.data.mirrors”
+
+      local idx = function (t, char)
+        local c = chardata [char]
+        if c then
+          local m = c [field] --> e.g. “mirror”
+          if m then
+            t [char] = m
+            return m
+          end
+        end
+        newspecial [char] = false
+        return char
+      end
+
+      setmetatableindex (newspecial, idx)
+    end
+
+  end
+
+  local mkcategories = function (characters) -- different from the others
+
+    local chardata         = characters.data
+    local categories       = characters.categories or { }
+    characters.categories  = categories
+
+    setmetatable (categories, { __index = function (t, char)
+      if char then
+        local c = chardata [char]
+        c = c.category or char
+        t [char] = c
+        return c
+      end
+    end})
+
+  end
+
+  local load_failed = false
+  local chardata --> characters.data; loaded on demand
+
+  local load_chardef = function ()
+
+    logreport ("both", 1, "aux", "Loading character metadata from %s.", chardef)
+    chardata = dofile (kpse.find_file (chardef, "lua"))
+
+    if chardata == nil then
+      logreport ("both", 0, "aux",
+                 "Could not load %s; continuing with empty character table.",
+                 chardef)
+      chardata    = { }
+      load_failed = true
+    end
+
+    characters             = { } --- nuke metatable
+    characters.data        = chardata
+    characters.classifiers = chardata.classifiers
+    chardata.classifiers   = nil
+
+    --- institute some of the functionality from char-ini.lua
+
+    mkcharspecial (characters, "mirrors",     "mirror")
+    mkcharspecial (characters, "directions",  "direction")
+    mkcharspecial (characters, "textclasses", "textclass")
+    mkcategories  (characters)
+
+  end
+
+  local charindex = function (t, k)
+    if chardata == nil and load_failed ~= true then
+      load_chardef ()
+    end
+
+    return characters [k]
+  end
+
+  setmetatableindex (characters, charindex)
+
+end
+
 local init_main = function ()
 
   local load_fontloader_module = luaotfload.loaders.fontloader
@@ -338,9 +469,6 @@ local init_main = function ()
                                         or "reference"
   fontloader = tostring (fontloader)
 
-  --- Preload the characters table. This may vanish later.
-  characters = luaotfload.loaders.luaotfload "characters"
-
   if fontloader == "reference" then
     logreport ("log", 0, "init", "Using reference fontloader.")
     load_fontloader_module (luaotfload.fontloader_package)