From 75c791b4fea141417fa088b9f5b9861ef57f4790 Mon Sep 17 00:00:00 2001
From: Albert Krewinkel <albert+github@zeitkraut.de>
Date: Sat, 9 Feb 2019 22:56:49 +0100
Subject: [PATCH] Lua filters: load module `pandoc` before calling `init.lua`
 (#5287)

The file `init.lua` in pandoc's data directory is run as part of
pandoc's Lua initialization process. Previously, the `pandoc` module was
loaded in `init.lua`, and the structure for marshaling was set-up after.
This allowed simple patching of element marshaling, but made using
`init.lua` more difficult:

  - it encouraged mixing essential initialization with user-defined
    customization;

  - upstream changes to init.lua had to be merged manually;

  - accidentally breaking marshaling by removing required modules was
    possible;

Instead, all required modules are now loaded before calling `init.lua`.
The file can be used entirely for user customization. Patching
marshaling functions, while discouraged, is still possible via the
`debug` module.
---
 data/init.lua               |  2 --
 doc/lua-filters.md          | 24 ++++++++++--------------
 src/Text/Pandoc/Lua/Init.hs | 33 +++++++++++++++++++++++++++------
 3 files changed, 37 insertions(+), 22 deletions(-)

diff --git a/data/init.lua b/data/init.lua
index 96813a7d1..84d826e73 100644
--- a/data/init.lua
+++ b/data/init.lua
@@ -1,5 +1,3 @@
 -- This Lua script is run every time the Lua interpreter is started when running
 -- a Lua filter. It can be customized to load additional modules or to alter the
 -- default modules.
-
-pandoc = require 'pandoc'
diff --git a/doc/lua-filters.md b/doc/lua-filters.md
index 01d2b1a87..09116cc16 100644
--- a/doc/lua-filters.md
+++ b/doc/lua-filters.md
@@ -219,21 +219,15 @@ Some pandoc functions have been made available in lua:
 
 # Lua interpreter initialization
 
-The way the Lua interpreter is set-up can be controlled by
-placing a file `init.lua` in pandoc's data directory. The
-default init file loads the `pandoc` and `pandoc.mediabag`
-modules:
+Initialization of pandoc's Lua interpreter can be controlled by
+placing a file `init.lua` in pandoc's data directory. A common
+use-case would be to load additional modules, or even to alter
+default modules.
 
-``` {.lua}
-pandoc = require 'pandoc'
-pandoc.mediabag = require 'pandoc.mediabag'
-```
-
-A common use-case would be to add code to load additional
-modules or to alter default modules. E.g., the following snippet
-adds all unicode-aware functions defined in the [`text`
-module](#module-text) to the default `string` module, prefixed
-with the string `uc_`.
+The following snippet is an example of code that might be useful
+when added to `init.lua`. The snippet adds all unicode-aware
+functions defined in the [`text` module] to the default `string`
+module, prefixed with the string `uc_`.
 
 ``` {.lua}
 for name, fn in pairs(require 'text') do
@@ -244,6 +238,8 @@ end
 This makes it possible to apply these functions on strings using
 colon syntax (`mystring:uc_upper()`).
 
+[`text` module]: #module-text
+
 # Examples
 
 The following filters are presented as examples.
diff --git a/src/Text/Pandoc/Lua/Init.hs b/src/Text/Pandoc/Lua/Init.hs
index c3096f611..f05076b20 100644
--- a/src/Text/Pandoc/Lua/Init.hs
+++ b/src/Text/Pandoc/Lua/Init.hs
@@ -49,6 +49,7 @@ import Text.Pandoc.Lua.Util (loadScriptFromDataDir)
 import qualified Foreign.Lua as Lua
 import qualified Foreign.Lua.Module.Text as Lua
 import qualified Text.Pandoc.Definition as Pandoc
+import qualified Text.Pandoc.Lua.Module.Pandoc as ModulePandoc
 
 -- | Lua error message
 newtype LuaException = LuaException String deriving (Show)
@@ -95,16 +96,37 @@ luaPackageParams = do
 
 -- | Initialize the lua state with all required values
 initLuaState :: LuaPackageParams -> Lua ()
-initLuaState luaPkgParams = do
+initLuaState pkgParams = do
   Lua.openlibs
   Lua.preloadTextModule "text"
-  installPandocPackageSearcher luaPkgParams
-  loadScriptFromDataDir (luaPkgDataDir luaPkgParams) "init.lua"
-  putConstructorsInRegistry
+  installPandocPackageSearcher pkgParams
+  initPandocModule
+  loadScriptFromDataDir (luaPkgDataDir pkgParams) "init.lua"
+ where
+  initPandocModule :: Lua ()
+  initPandocModule = do
+    -- Push module table
+    ModulePandoc.pushModule (luaPkgDataDir pkgParams)
+    -- register as loaded module
+    Lua.pushvalue Lua.stackTop
+    Lua.getfield Lua.registryindex Lua.loadedTableRegistryField
+    Lua.setfield (Lua.nthFromTop 2) "pandoc"
+    Lua.pop 1
+    -- copy constructors into registry
+    putConstructorsInRegistry
+    -- assign module to global variable
+    Lua.setglobal "pandoc"
 
+-- | AST elements are marshaled via normal constructor functions in the
+-- @pandoc@ module. However, accessing Lua globals from Haskell is
+-- expensive (due to error handling). Accessing the Lua registry is much
+-- cheaper, which is why the constructor functions are copied into the
+-- Lua registry and called from there.
+--
+-- This function expects the @pandoc@ module to be at the top of the
+-- stack.
 putConstructorsInRegistry :: Lua ()
 putConstructorsInRegistry = do
-  Lua.getglobal "pandoc"
   constrsToReg $ Pandoc.Pandoc mempty mempty
   constrsToReg $ Pandoc.Str mempty
   constrsToReg $ Pandoc.Para mempty
@@ -113,7 +135,6 @@ putConstructorsInRegistry = do
   constrsToReg $ Pandoc.Citation mempty mempty mempty Pandoc.AuthorInText 0 0
   putInReg "Attr"  -- used for Attr type alias
   putInReg "ListAttributes"  -- used for ListAttributes type alias
-  Lua.pop 1
  where
   constrsToReg :: Data a => a -> Lua ()
   constrsToReg = mapM_ (putInReg . showConstr) . dataTypeConstrs . dataTypeOf