Add support for compiling to wasm.

This builds on the excellent work by @TerrorJack at
haskell-wasm/pandoc-wasm.

However, we can now make use of a new http cabal flag that simplify
things, in addition to a small API change (FromJSON for LogMessage).

The interface of `wasm_main` now taes a JSON-encoded Opts as input,
instead of a command-line argument string.

Input/output files can now be specified (these will have to come
from the virtual file system).

The Makefile has a new pandoc.wasm target. Note that the
environment variable `OPTIMIZE_WASM` may be set to 0 for faster builds.

Log messages and errors are rendered to `/stderr`. Regular
output goes to `/stdout`, and input is taken from `/stdin`.
Another file `/warnings` collects warnings (using `--log-file`);
these are then rendered to `/stderr`.

Author: jgm

Makefile

@@ -17,6 +17,7 @@ WEBSITE=../../web/pandoc.org
 REVISION?=1
 BENCHARGS?=--csv bench_$(TIMESTAMP).csv $(BASELINECMD) --timeout=6 +RTS -T --nonmoving-gc -RTS $(if $(PATTERN),--pattern "$(PATTERN)",)
 pandoc=$(shell cabal list-bin $(CABALOPTS) pandoc-cli)
+OPTIMIZE_WASM?=1
 
 all: build test binpath ## build executable and run tests
 .PHONY: all
@@ -326,3 +327,11 @@ release-checklist-$(VERSION).org: RELEASE-CHECKLIST-TEMPLATE.org
 hie.yaml: ## regenerate hie.yaml
 	gen-hie > $@
 .PHONY: hie.yaml
+
+pandoc.wasm:
+	rm $@
+ifeq ($(OPTIMIZE_WASM),1)
+	echo 'wasm32-wasi-cabal build pandoc-cli && wasm-opt --low-memory-unused --converge --gufa --flatten --rereloop -Oz $$(wasm32-wasi-cabal list-bin pandoc-cli) -o $@' | nix shell 'gitlab:haskell-wasm/ghc-wasm-meta?host=gitlab.haskell.org'
+else
+	echo 'wasm32-wasi-cabal build pandoc-cli && cp $$(wasm32-wasi-cabal list-bin pandoc-cli) $@' | nix shell 'gitlab:haskell-wasm/ghc-wasm-meta?host=gitlab.haskell.org'
+endif

cabal.project

@@ -2,11 +2,12 @@ packages: .
           pandoc-lua-engine
           pandoc-server
           pandoc-cli
-tests: True
-flags: +embed_data_files
 constraints: skylighting-format-blaze-html >= 0.1.1.3,
              skylighting-format-context >= 0.1.0.2
 
+package pandoc
+  flags: +embed_data_files +http
+
 source-repository-package
   type: git
   location: https://github.com/jgm/asciidoc-hs.git
@@ -31,3 +32,66 @@ source-repository-package
   type: git
   location: https://github.com/jgm/djoths.git
   tag: a6fc625fd58aa39df05dd1d82ee005e681815095
+
+if arch(wasm32)
+  tests: False
+
+  package aeson
+    flags: -ordered-keymap
+
+  package crypton
+    ghc-options: -optc-DARGON2_NO_THREADS
+
+  package digest
+    flags: -pkg-config
+
+  package pandoc
+    flags: +embed_data_files -http
+
+  package pandoc-cli
+    flags: -lua -server
+
+  allow-newer:
+    all:base,
+    all:binary,
+    all:bytestring,
+    all:containers,
+    all:ghc-bignum,
+    all:template-haskell,
+    all:text,
+    all:time
+
+  source-repository-package
+    type: git
+    location: https://github.com/haskell-wasm/conduit.git
+    tag: ff33329247f2ef321dcab836e98c1bcfaff2bd13
+    subdir: conduit-extra
+
+  source-repository-package
+    type: git
+    location: https://github.com/haskell-wasm/foundation.git
+    tag: 8e6dd48527fb429c1922083a5030ef88e3d58dd3
+    subdir: basement
+
+  source-repository-package
+    type: git
+    location: https://github.com/haskell-wasm/hs-memory.git
+    tag: a198a76c584dc2cfdcde6b431968de92a5fed65e
+
+  source-repository-package
+    type: git
+    location: https://github.com/haskell-wasm/streaming-commons.git
+    tag: 7e9c38b2fd55ce50d3f74fe708ca47db8c9bb315
+
+  source-repository-package
+    type: git
+    location: https://github.com/haskell-wasm/xml.git
+    tag: bc793dc9bc29c92245d3482a54d326abd3ae1403
+    subdir: xml-conduit
+
+  -- https://github.com/haskellari/splitmix/pull/73
+  source-repository-package
+    type: git
+    location: https://github.com/amesgen/splitmix
+    tag: 5f5b766d97dc735ac228215d240a3bb90bc2ff75
+

pandoc-cli/pandoc-cli.cabal

@@ -61,7 +61,7 @@ common common-options
 
 common common-executable
   import:           common-options
-  ghc-options:      -rtsopts -with-rtsopts=-A8m -threaded
+  ghc-options:      -rtsopts -with-rtsopts=-H64m
 
 executable pandoc
   import:          common-executable
@@ -74,6 +74,11 @@ executable pandoc
                    text
   other-modules:   PandocCLI.Lua
                  , PandocCLI.Server
+
+  if arch(wasm32)
+    build-depends: aeson, bytestring
+    ghc-options:    -optl-Wl,--export=__wasm_call_ctors,--export=hs_init_with_rtsopts,--export=malloc,--export=wasm_main
+
   if flag(nightly)
     cpp-options:    -DNIGHTLY
     build-depends:  template-haskell,

pandoc-cli/src/pandoc.hs

@@ -1,5 +1,7 @@
 {-# LANGUAGE CPP #-}
+{-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE TemplateHaskell #-}
+
 {- |
    Module      : Main
    Copyright   : Copyright (C) 2006-2024 John MacFarlane
@@ -15,7 +17,8 @@ writers.
 module Main where
 import qualified Control.Exception as E
 import System.Environment (getArgs, getProgName)
-import Text.Pandoc.App ( convertWithOpts, defaultOpts, options
+import Data.Maybe (fromMaybe)
+import Text.Pandoc.App ( convertWithOpts, defaultOpts, options, Opt(..)
                        , parseOptionsFromArgs, handleOptInfo, versionInfo )
 import Text.Pandoc.Error (handleError)
 import Data.Monoid (Any(..))
@@ -29,6 +32,20 @@ import qualified Language.Haskell.TH as TH
 import Data.Time
 #endif
 
+#if defined(wasm32_HOST_ARCH)
+import Control.Exception
+import Foreign
+import Foreign.C
+import qualified Data.Aeson as Aeson
+import qualified Text.Pandoc.UTF8 as UTF8
+import qualified Data.Text.IO as TIO
+import qualified Data.ByteString.Lazy as BL
+import Text.Pandoc.Error (renderError, PandocError)
+import Text.Pandoc.Logging (showLogMessage)
+#else
+import System.Exit
+#endif
+
 #ifdef NIGHTLY
 versionSuffix :: String
 versionSuffix = "-nightly-" ++
@@ -39,6 +56,35 @@ versionSuffix :: String
 versionSuffix = ""
 #endif
 
+#if defined(wasm32_HOST_ARCH)
+
+foreign export ccall "wasm_main" wasm_main :: Ptr CChar -> Int -> IO ()
+
+wasm_main :: Ptr CChar -> Int -> IO ()
+wasm_main raw_args_ptr raw_args_len =
+  catch act (\(err :: SomeException) -> writeFile "/stderr" (show err))
+  where
+    act = do
+      args <- peekCStringLen (raw_args_ptr, raw_args_len)
+      free raw_args_ptr
+      engine <- getEngine
+      let aesonRes = Aeson.eitherDecode (UTF8.fromStringLazy args)
+      case aesonRes of
+        Left e -> error e
+        Right opts -> do
+          let opts' = opts{ optInputFiles = Just $ fromMaybe ["/stdin"] (optInputFiles opts)
+                          , optOutputFile = Just $ fromMaybe "/stdout" (optOutputFile opts)
+                          , optLogFile = Just $ fromMaybe "/warnings" (optLogFile opts)
+                          }
+          E.catch (convertWithOpts engine opts') $
+            \(e :: PandocError) -> TIO.writeFile "/stderr" (renderError e)
+          res <- Aeson.eitherDecode <$> BL.readFile "/warnings"
+          case res of
+            Left e -> writeFile "/stderr" e
+            Right msgs -> TIO.writeFile "/stderr" (T.unlines $ map showLogMessage msgs)
+
+#endif
+
 main :: IO ()
 main = E.handle (handleError . Left) $ do
   prg <- getProgName