pandoc/src/pandoc.hs

889 lines
37 KiB
Haskell
Raw Normal View History

{-
2011-02-07 02:04:19 +01:00
Copyright (C) 2006-2011 John MacFarlane <jgm@berkeley.edu>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
-}
{- |
Module : Main
2011-02-07 02:04:19 +01:00
Copyright : Copyright (C) 2006-2011 John MacFarlane
License : GNU GPL, version 2 or above
Maintainer : John MacFarlane <jgm@berkeley@edu>
Stability : alpha
Portability : portable
Parses command-line options and calls the appropriate readers and
writers.
-}
module Main where
import Text.Pandoc
import Text.Pandoc.S5 (s5HeaderIncludes)
import Text.Pandoc.Shared ( tabFilter, ObfuscationMethod (..), readDataFile,
2011-01-21 07:48:20 +01:00
headerShift, findDataFile, normalize )
#ifdef _HIGHLIGHTING
import Text.Pandoc.Highlighting ( languages )
#endif
import System.Environment ( getArgs, getProgName )
import System.Exit ( exitWith, ExitCode (..) )
import System.FilePath
import System.Console.GetOpt
import Data.Char ( toLower )
import Data.List ( intercalate, isSuffixOf, isPrefixOf )
import System.Directory ( getAppUserDataDirectory, doesFileExist )
import System.IO ( stdout, stderr )
import System.IO.Error ( isDoesNotExistError )
import Control.Exception.Extensible ( throwIO )
import qualified Text.Pandoc.UTF8 as UTF8
import Text.CSL
import Text.Pandoc.Biblio
import Control.Monad (when, unless, liftM)
import Network.HTTP (simpleHTTP, mkRequest, getResponseBody, RequestMethod(..))
import Network.URI (parseURI, isURI, URI(..))
import qualified Data.ByteString.Lazy as B
import Data.ByteString.Lazy.UTF8 (toString )
import Codec.Binary.UTF8.String (decodeString, encodeString)
copyrightMessage :: String
2011-02-07 02:04:19 +01:00
copyrightMessage = "\nCopyright (C) 2006-2011 John MacFarlane\n" ++
"Web: http://johnmacfarlane.net/pandoc\n" ++
"This is free software; see the source for copying conditions. There is no\n" ++
"warranty, not even for merchantability or fitness for a particular purpose."
compileInfo :: String
compileInfo =
"\nCompiled with citeproc support." ++
#ifdef _HIGHLIGHTING
"\nCompiled with syntax highlighting support for:\n" ++
wrapWords 78 languages ++
#endif
""
-- | Converts a list of strings into a single string with the items printed as
-- comma separated words in lines with a maximum line length.
wrapWords :: Int -> [String] -> String
wrapWords c = wrap' c c where
wrap' _ _ [] = ""
wrap' cols remaining (x:xs) = if remaining == cols
then x ++ wrap' cols (remaining - length x) xs
else if (length x + 1) > remaining
then ",\n" ++ x ++ wrap' cols (cols - length x) xs
else ", " ++ x ++ wrap' cols (remaining - (length x + 2)) xs
isNonTextOutput :: String -> Bool
isNonTextOutput = (`elem` ["odt","epub"])
-- | Data structure for command line options.
data Opt = Opt
{ optTabStop :: Int -- ^ Number of spaces per tab
, optPreserveTabs :: Bool -- ^ Preserve tabs instead of converting to spaces
, optStandalone :: Bool -- ^ Include header, footer
, optReader :: String -- ^ Reader format
, optWriter :: String -- ^ Writer format
, optParseRaw :: Bool -- ^ Parse unconvertable HTML and TeX
, optTableOfContents :: Bool -- ^ Include table of contents
, optTransforms :: [Pandoc -> Pandoc] -- ^ Doc transforms to apply
, optTemplate :: Maybe FilePath -- ^ Custom template
, optVariables :: [(String,String)] -- ^ Template variables to set
, optOutputFile :: String -- ^ Name of output file
, optNumberSections :: Bool -- ^ Number sections in LaTeX
, optSectionDivs :: Bool -- ^ Put sections in div tags in HTML
, optIncremental :: Bool -- ^ Use incremental lists in Slidy/S5
, optOffline :: Bool -- ^ Make slideshow accessible offline
, optXeTeX :: Bool -- ^ Format latex for xetex
, optSmart :: Bool -- ^ Use smart typography
, optHtml5 :: Bool -- ^ Produce HTML5 in HTML
, optChapters :: Bool -- ^ Use chapter for top-level sects
, optHTMLMathMethod :: HTMLMathMethod -- ^ Method to print HTML math
, optReferenceODT :: Maybe FilePath -- ^ Path of reference.odt
, optEPUBStylesheet :: Maybe String -- ^ EPUB stylesheet
, optEPUBMetadata :: String -- ^ EPUB metadata
, optDumpArgs :: Bool -- ^ Output command-line arguments
, optIgnoreArgs :: Bool -- ^ Ignore command-line arguments
, optStrict :: Bool -- ^ Use strict markdown syntax
Extensive changes stemming from a rethinking of the Pandoc data structure. Key and Note blocks have been removed. Link and image URLs are now stored directly in Link and Image inlines, and note blocks are stored in Note inlines. This requires changes in both parsers and writers. Markdown and RST parsers need to extract data from key and note blocks and insert them into the relevant inline elements. Other parsers can be simplified, since there is no longer any need to construct separate key and note blocks. Markdown, RST, and HTML writers need to construct lists of notes; Markdown and RST writers need to construct lists of link references (when the --reference-links option is specified); and the RST writer needs to construct a list of image substitution references. All writers have been rewritten to use the State monad when state is required. This rewrite yields a small speed boost and considerably cleaner code. * Text/Pandoc/Definition.hs: + blocks: removed Key and Note + inlines: removed NoteRef, added Note + modified Target: there is no longer a 'Ref' target; all targets are explicit URL, title pairs * Text/Pandoc/Shared.hs: + Added 'Reference', 'isNoteBlock', 'isKeyBlock', 'isLineClump', used in some of the readers. + Removed 'generateReference', 'keyTable', 'replaceReferenceLinks', 'replaceRefLinksBlockList', along with some auxiliary functions used only by them. These are no longer needed, since reference links are resolved in the Markdown and RST readers. + Moved 'inTags', 'selfClosingTag', 'inTagsSimple', and 'inTagsIndented' to the Docbook writer, since that is now the only module that uses them. + Changed name of 'escapeSGMLString' to 'escapeStringForXML' + Added KeyTable and NoteTable types + Removed fields from ParserState; 'stateKeyBlocks', 'stateKeysUsed', 'stateNoteBlocks', 'stateNoteIdentifiers', 'stateInlineLinks'. Added 'stateKeys' and 'stateNotes'. + Added clause for Note to 'prettyBlock'. + Added 'writerNotes', 'writerReferenceLinks' fields to WriterOptions. * Text/Pandoc/Entities.hs: Renamed 'escapeSGMLChar' and 'escapeSGMLString' to 'escapeCharForXML' and 'escapeStringForXML' * Text/ParserCombinators/Pandoc.hs: Added lineClump parser: parses a raw line block up to and including following blank lines. * Main.hs: Replaced --inline-links with --reference-links. * README: + Documented --reference-links and removed description of --inline-links. + Added note that footnotes may occur anywhere in the document, but must be at the outer level, not embedded in block elements. * man/man1/pandoc.1, man/man1/html2markdown.1: Removed --inline-links option, added --reference-links option * Markdown and RST readers: + Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, all note and key blocks are parsed on a first pass through the document. Once tables of notes and keys have been constructed, the remaining parts of the document are reassembled and parsed. + Refactored link parsers. * LaTeX and HTML readers: Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, notes and references can be parsed in a single pass through the document. * RST, Markdown, and HTML writers: Rewrote using state monad new Pandoc and definition. State is used to hold lists of references footnotes to and be printed at the end of the document. * RTF and LaTeX writers: Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Docbook writer: + Moved the functions 'attributeList', 'inTags', 'selfClosingTag', 'inTagsSimple', 'inTagsIndented' from Text/Pandoc/Shared, since they are now used only by the Docbook writer. + Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Updated test suite * Throughout: old haskell98 module names replaced by hierarchical module names, e.g. List by Data.List. * debian/control: Include libghc6-xhtml-dev instead of libghc6-html-dev in "Build-Depends." * cabalize: + Remove haskell98 from BASE_DEPENDS (since now the new hierarchical module names are being used throughout) + Added mtl to BASE_DEPENDS (needed for state monad) + Removed html from GHC66_DEPENDS (not needed since xhtml is now used) git-svn-id: https://pandoc.googlecode.com/svn/trunk@580 788f1e2b-df1e-0410-8736-df70ead52e1b
2007-04-10 03:56:50 +02:00
, optReferenceLinks :: Bool -- ^ Use reference links in writing markdown, rst
, optWrapText :: Bool -- ^ Wrap text
, optColumns :: Int -- ^ Line length in characters
, optPlugins :: [Pandoc -> IO Pandoc] -- ^ Plugins to apply
, optEmailObfuscation :: ObfuscationMethod
, optIdentifierPrefix :: String
, optIndentedCodeClasses :: [String] -- ^ Default classes for indented code blocks
, optDataDir :: Maybe FilePath
, optCiteMethod :: CiteMethod -- ^ Method to output cites
, optBibliography :: [String]
, optCslFile :: FilePath
, optListings :: Bool -- ^ Use listings package for code blocks
, optAscii :: Bool -- ^ Avoid using nonascii characters
}
-- | Defaults for command-line options.
defaultOpts :: Opt
defaultOpts = Opt
{ optTabStop = 4
, optPreserveTabs = False
, optStandalone = False
, optReader = "" -- null for default reader
, optWriter = "" -- null for default writer
, optParseRaw = False
, optTableOfContents = False
, optTransforms = []
, optTemplate = Nothing
, optVariables = []
, optOutputFile = "-" -- "-" means stdout
, optNumberSections = False
, optSectionDivs = False
, optIncremental = False
, optOffline = False
, optXeTeX = False
, optSmart = False
, optHtml5 = False
, optChapters = False
, optHTMLMathMethod = PlainMath
, optReferenceODT = Nothing
, optEPUBStylesheet = Nothing
, optEPUBMetadata = ""
, optDumpArgs = False
, optIgnoreArgs = False
, optStrict = False
Extensive changes stemming from a rethinking of the Pandoc data structure. Key and Note blocks have been removed. Link and image URLs are now stored directly in Link and Image inlines, and note blocks are stored in Note inlines. This requires changes in both parsers and writers. Markdown and RST parsers need to extract data from key and note blocks and insert them into the relevant inline elements. Other parsers can be simplified, since there is no longer any need to construct separate key and note blocks. Markdown, RST, and HTML writers need to construct lists of notes; Markdown and RST writers need to construct lists of link references (when the --reference-links option is specified); and the RST writer needs to construct a list of image substitution references. All writers have been rewritten to use the State monad when state is required. This rewrite yields a small speed boost and considerably cleaner code. * Text/Pandoc/Definition.hs: + blocks: removed Key and Note + inlines: removed NoteRef, added Note + modified Target: there is no longer a 'Ref' target; all targets are explicit URL, title pairs * Text/Pandoc/Shared.hs: + Added 'Reference', 'isNoteBlock', 'isKeyBlock', 'isLineClump', used in some of the readers. + Removed 'generateReference', 'keyTable', 'replaceReferenceLinks', 'replaceRefLinksBlockList', along with some auxiliary functions used only by them. These are no longer needed, since reference links are resolved in the Markdown and RST readers. + Moved 'inTags', 'selfClosingTag', 'inTagsSimple', and 'inTagsIndented' to the Docbook writer, since that is now the only module that uses them. + Changed name of 'escapeSGMLString' to 'escapeStringForXML' + Added KeyTable and NoteTable types + Removed fields from ParserState; 'stateKeyBlocks', 'stateKeysUsed', 'stateNoteBlocks', 'stateNoteIdentifiers', 'stateInlineLinks'. Added 'stateKeys' and 'stateNotes'. + Added clause for Note to 'prettyBlock'. + Added 'writerNotes', 'writerReferenceLinks' fields to WriterOptions. * Text/Pandoc/Entities.hs: Renamed 'escapeSGMLChar' and 'escapeSGMLString' to 'escapeCharForXML' and 'escapeStringForXML' * Text/ParserCombinators/Pandoc.hs: Added lineClump parser: parses a raw line block up to and including following blank lines. * Main.hs: Replaced --inline-links with --reference-links. * README: + Documented --reference-links and removed description of --inline-links. + Added note that footnotes may occur anywhere in the document, but must be at the outer level, not embedded in block elements. * man/man1/pandoc.1, man/man1/html2markdown.1: Removed --inline-links option, added --reference-links option * Markdown and RST readers: + Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, all note and key blocks are parsed on a first pass through the document. Once tables of notes and keys have been constructed, the remaining parts of the document are reassembled and parsed. + Refactored link parsers. * LaTeX and HTML readers: Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, notes and references can be parsed in a single pass through the document. * RST, Markdown, and HTML writers: Rewrote using state monad new Pandoc and definition. State is used to hold lists of references footnotes to and be printed at the end of the document. * RTF and LaTeX writers: Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Docbook writer: + Moved the functions 'attributeList', 'inTags', 'selfClosingTag', 'inTagsSimple', 'inTagsIndented' from Text/Pandoc/Shared, since they are now used only by the Docbook writer. + Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Updated test suite * Throughout: old haskell98 module names replaced by hierarchical module names, e.g. List by Data.List. * debian/control: Include libghc6-xhtml-dev instead of libghc6-html-dev in "Build-Depends." * cabalize: + Remove haskell98 from BASE_DEPENDS (since now the new hierarchical module names are being used throughout) + Added mtl to BASE_DEPENDS (needed for state monad) + Removed html from GHC66_DEPENDS (not needed since xhtml is now used) git-svn-id: https://pandoc.googlecode.com/svn/trunk@580 788f1e2b-df1e-0410-8736-df70ead52e1b
2007-04-10 03:56:50 +02:00
, optReferenceLinks = False
, optWrapText = True
, optColumns = 72
, optPlugins = []
, optEmailObfuscation = JavascriptObfuscation
, optIdentifierPrefix = ""
, optIndentedCodeClasses = []
, optDataDir = Nothing
, optCiteMethod = Citeproc
, optBibliography = []
, optCslFile = ""
, optListings = False
, optAscii = False
}
-- | A list of functions, each transforming the options data structure
-- in response to a command-line option.
options :: [OptDescr (Opt -> IO Opt)]
options =
[ Option "fr" ["from","read"]
(ReqArg
(\arg opt -> return opt { optReader = map toLower arg })
"FORMAT")
""
, Option "tw" ["to","write"]
(ReqArg
(\arg opt -> return opt { optWriter = map toLower arg })
"FORMAT")
""
, Option "s" ["standalone"]
(NoArg
(\opt -> return opt { optStandalone = True }))
"" -- "Include needed header and footer on output"
, Option "o" ["output"]
(ReqArg
(\arg opt -> return opt { optOutputFile = arg })
"FILENAME")
"" -- "Name of output file"
, Option "p" ["preserve-tabs"]
(NoArg
(\opt -> return opt { optPreserveTabs = True }))
"" -- "Preserve tabs instead of converting to spaces"
, Option "" ["tab-stop"]
(ReqArg
(\arg opt ->
case reads arg of
[(t,"")] | t > 0 -> return opt { optTabStop = t }
_ -> do
UTF8.hPutStrLn stderr $
"tab-stop must be a number greater than 0"
exitWith $ ExitFailure 31)
"NUMBER")
"" -- "Tab stop (default 4)"
, Option "" ["strict"]
(NoArg
(\opt -> return opt { optStrict = True } ))
"" -- "Disable markdown syntax extensions"
2011-01-21 07:48:20 +01:00
, Option "" ["normalize"]
(NoArg
(\opt -> return opt { optTransforms =
normalize : optTransforms opt } ))
"" -- "Normalize the Pandoc AST"
Extensive changes stemming from a rethinking of the Pandoc data structure. Key and Note blocks have been removed. Link and image URLs are now stored directly in Link and Image inlines, and note blocks are stored in Note inlines. This requires changes in both parsers and writers. Markdown and RST parsers need to extract data from key and note blocks and insert them into the relevant inline elements. Other parsers can be simplified, since there is no longer any need to construct separate key and note blocks. Markdown, RST, and HTML writers need to construct lists of notes; Markdown and RST writers need to construct lists of link references (when the --reference-links option is specified); and the RST writer needs to construct a list of image substitution references. All writers have been rewritten to use the State monad when state is required. This rewrite yields a small speed boost and considerably cleaner code. * Text/Pandoc/Definition.hs: + blocks: removed Key and Note + inlines: removed NoteRef, added Note + modified Target: there is no longer a 'Ref' target; all targets are explicit URL, title pairs * Text/Pandoc/Shared.hs: + Added 'Reference', 'isNoteBlock', 'isKeyBlock', 'isLineClump', used in some of the readers. + Removed 'generateReference', 'keyTable', 'replaceReferenceLinks', 'replaceRefLinksBlockList', along with some auxiliary functions used only by them. These are no longer needed, since reference links are resolved in the Markdown and RST readers. + Moved 'inTags', 'selfClosingTag', 'inTagsSimple', and 'inTagsIndented' to the Docbook writer, since that is now the only module that uses them. + Changed name of 'escapeSGMLString' to 'escapeStringForXML' + Added KeyTable and NoteTable types + Removed fields from ParserState; 'stateKeyBlocks', 'stateKeysUsed', 'stateNoteBlocks', 'stateNoteIdentifiers', 'stateInlineLinks'. Added 'stateKeys' and 'stateNotes'. + Added clause for Note to 'prettyBlock'. + Added 'writerNotes', 'writerReferenceLinks' fields to WriterOptions. * Text/Pandoc/Entities.hs: Renamed 'escapeSGMLChar' and 'escapeSGMLString' to 'escapeCharForXML' and 'escapeStringForXML' * Text/ParserCombinators/Pandoc.hs: Added lineClump parser: parses a raw line block up to and including following blank lines. * Main.hs: Replaced --inline-links with --reference-links. * README: + Documented --reference-links and removed description of --inline-links. + Added note that footnotes may occur anywhere in the document, but must be at the outer level, not embedded in block elements. * man/man1/pandoc.1, man/man1/html2markdown.1: Removed --inline-links option, added --reference-links option * Markdown and RST readers: + Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, all note and key blocks are parsed on a first pass through the document. Once tables of notes and keys have been constructed, the remaining parts of the document are reassembled and parsed. + Refactored link parsers. * LaTeX and HTML readers: Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, notes and references can be parsed in a single pass through the document. * RST, Markdown, and HTML writers: Rewrote using state monad new Pandoc and definition. State is used to hold lists of references footnotes to and be printed at the end of the document. * RTF and LaTeX writers: Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Docbook writer: + Moved the functions 'attributeList', 'inTags', 'selfClosingTag', 'inTagsSimple', 'inTagsIndented' from Text/Pandoc/Shared, since they are now used only by the Docbook writer. + Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Updated test suite * Throughout: old haskell98 module names replaced by hierarchical module names, e.g. List by Data.List. * debian/control: Include libghc6-xhtml-dev instead of libghc6-html-dev in "Build-Depends." * cabalize: + Remove haskell98 from BASE_DEPENDS (since now the new hierarchical module names are being used throughout) + Added mtl to BASE_DEPENDS (needed for state monad) + Removed html from GHC66_DEPENDS (not needed since xhtml is now used) git-svn-id: https://pandoc.googlecode.com/svn/trunk@580 788f1e2b-df1e-0410-8736-df70ead52e1b
2007-04-10 03:56:50 +02:00
, Option "" ["reference-links"]
(NoArg
Extensive changes stemming from a rethinking of the Pandoc data structure. Key and Note blocks have been removed. Link and image URLs are now stored directly in Link and Image inlines, and note blocks are stored in Note inlines. This requires changes in both parsers and writers. Markdown and RST parsers need to extract data from key and note blocks and insert them into the relevant inline elements. Other parsers can be simplified, since there is no longer any need to construct separate key and note blocks. Markdown, RST, and HTML writers need to construct lists of notes; Markdown and RST writers need to construct lists of link references (when the --reference-links option is specified); and the RST writer needs to construct a list of image substitution references. All writers have been rewritten to use the State monad when state is required. This rewrite yields a small speed boost and considerably cleaner code. * Text/Pandoc/Definition.hs: + blocks: removed Key and Note + inlines: removed NoteRef, added Note + modified Target: there is no longer a 'Ref' target; all targets are explicit URL, title pairs * Text/Pandoc/Shared.hs: + Added 'Reference', 'isNoteBlock', 'isKeyBlock', 'isLineClump', used in some of the readers. + Removed 'generateReference', 'keyTable', 'replaceReferenceLinks', 'replaceRefLinksBlockList', along with some auxiliary functions used only by them. These are no longer needed, since reference links are resolved in the Markdown and RST readers. + Moved 'inTags', 'selfClosingTag', 'inTagsSimple', and 'inTagsIndented' to the Docbook writer, since that is now the only module that uses them. + Changed name of 'escapeSGMLString' to 'escapeStringForXML' + Added KeyTable and NoteTable types + Removed fields from ParserState; 'stateKeyBlocks', 'stateKeysUsed', 'stateNoteBlocks', 'stateNoteIdentifiers', 'stateInlineLinks'. Added 'stateKeys' and 'stateNotes'. + Added clause for Note to 'prettyBlock'. + Added 'writerNotes', 'writerReferenceLinks' fields to WriterOptions. * Text/Pandoc/Entities.hs: Renamed 'escapeSGMLChar' and 'escapeSGMLString' to 'escapeCharForXML' and 'escapeStringForXML' * Text/ParserCombinators/Pandoc.hs: Added lineClump parser: parses a raw line block up to and including following blank lines. * Main.hs: Replaced --inline-links with --reference-links. * README: + Documented --reference-links and removed description of --inline-links. + Added note that footnotes may occur anywhere in the document, but must be at the outer level, not embedded in block elements. * man/man1/pandoc.1, man/man1/html2markdown.1: Removed --inline-links option, added --reference-links option * Markdown and RST readers: + Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, all note and key blocks are parsed on a first pass through the document. Once tables of notes and keys have been constructed, the remaining parts of the document are reassembled and parsed. + Refactored link parsers. * LaTeX and HTML readers: Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, notes and references can be parsed in a single pass through the document. * RST, Markdown, and HTML writers: Rewrote using state monad new Pandoc and definition. State is used to hold lists of references footnotes to and be printed at the end of the document. * RTF and LaTeX writers: Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Docbook writer: + Moved the functions 'attributeList', 'inTags', 'selfClosingTag', 'inTagsSimple', 'inTagsIndented' from Text/Pandoc/Shared, since they are now used only by the Docbook writer. + Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Updated test suite * Throughout: old haskell98 module names replaced by hierarchical module names, e.g. List by Data.List. * debian/control: Include libghc6-xhtml-dev instead of libghc6-html-dev in "Build-Depends." * cabalize: + Remove haskell98 from BASE_DEPENDS (since now the new hierarchical module names are being used throughout) + Added mtl to BASE_DEPENDS (needed for state monad) + Removed html from GHC66_DEPENDS (not needed since xhtml is now used) git-svn-id: https://pandoc.googlecode.com/svn/trunk@580 788f1e2b-df1e-0410-8736-df70ead52e1b
2007-04-10 03:56:50 +02:00
(\opt -> return opt { optReferenceLinks = True } ))
"" -- "Use reference links in parsing HTML"
, Option "R" ["parse-raw"]
(NoArg
(\opt -> return opt { optParseRaw = True }))
"" -- "Parse untranslatable HTML codes and LaTeX environments as raw"
, Option "S" ["smart"]
(NoArg
(\opt -> return opt { optSmart = True }))
"" -- "Use smart quotes, dashes, and ellipses"
, Option "5" ["html5"]
(NoArg
(\opt -> return opt { optHtml5 = True }))
"" -- "Produce HTML5 in HTML output"
, Option "m" ["latexmathml", "asciimathml"]
(OptArg
(\arg opt ->
return opt { optHTMLMathMethod = LaTeXMathML arg })
"URL")
"" -- "Use LaTeXMathML script in html output"
, Option "" ["mathml"]
(OptArg
(\arg opt ->
return opt { optHTMLMathMethod = MathML arg })
"URL")
"" -- "Use mathml for HTML math"
, Option "" ["mimetex"]
(OptArg
(\arg opt -> do
let url' = case arg of
Just u -> u ++ "?"
Nothing -> "/cgi-bin/mimetex.cgi?"
return opt { optHTMLMathMethod = WebTeX url' })
"URL")
"" -- "Use mimetex for HTML math"
, Option "" ["webtex"]
(OptArg
(\arg opt -> do
let url' = case arg of
Just u -> u
Nothing -> "http://chart.apis.google.com/chart?cht=tx&chl="
return opt { optHTMLMathMethod = WebTeX url' })
"URL")
"" -- "Use web service for HTML math"
, Option "" ["jsmath"]
(OptArg
(\arg opt -> return opt { optHTMLMathMethod = JsMath arg})
"URL")
"" -- "Use jsMath for HTML math"
, Option "" ["mathjax"]
(ReqArg
(\arg opt -> return opt { optHTMLMathMethod = MathJax arg})
"URL")
"" -- "Use MathJax for HTML math"
, Option "" ["gladtex"]
(NoArg
(\opt -> return opt { optHTMLMathMethod = GladTeX }))
"" -- "Use gladtex for HTML math"
, Option "i" ["incremental"]
(NoArg
(\opt -> return opt { optIncremental = True }))
"" -- "Make list items display incrementally in Slidy/S5"
, Option "" ["offline"]
(NoArg
2010-07-24 09:49:10 +02:00
(\opt -> return opt { optOffline = True,
optStandalone = True }))
"" -- "Make slide shows include all the needed js and css"
, Option "" ["xetex"]
(NoArg
(\opt -> do
UTF8.hPutStrLn stderr $ "pandoc: --xetex is deprecated. "
++ "It is no longer needed for use with XeTeX."
return opt { optXeTeX = True }))
"" -- "Format latex for processing by XeTeX"
, Option "" ["chapters"]
(NoArg
(\opt -> return opt { optChapters = True }))
"" -- "Use chapter for top-level sections in LaTeX, DocBook"
, Option "N" ["number-sections"]
(NoArg
(\opt -> return opt { optNumberSections = True }))
"" -- "Number sections in LaTeX"
, Option "" ["listings"]
(NoArg
(\opt -> return opt { optListings = True }))
"" -- "Use listings package for LaTeX code blocks"
, Option "" ["section-divs"]
(NoArg
(\opt -> return opt { optSectionDivs = True }))
"" -- "Put sections in div tags in HTML"
, Option "" ["no-wrap"]
(NoArg
(\opt -> return opt { optWrapText = False }))
"" -- "Do not wrap text in output"
, Option "" ["columns"]
(ReqArg
(\arg opt ->
case reads arg of
[(t,"")] | t > 0 -> return opt { optColumns = t }
_ -> do
UTF8.hPutStrLn stderr $
"columns must be a number greater than 0"
exitWith $ ExitFailure 33)
"NUMBER")
"" -- "Length of line in characters"
, Option "" ["ascii"]
(NoArg
(\opt -> return opt { optAscii = True }))
"" -- "Avoid using non-ascii characters in output"
, Option "" ["email-obfuscation"]
(ReqArg
(\arg opt -> do
method <- case arg of
"references" -> return ReferenceObfuscation
"javascript" -> return JavascriptObfuscation
"none" -> return NoObfuscation
_ -> UTF8.hPutStrLn stderr ("Error: Unknown obfuscation method: " ++ arg) >>
exitWith (ExitFailure 6)
return opt { optEmailObfuscation = method })
"none|javascript|references")
"" -- "Method for obfuscating email in HTML"
, Option "" ["id-prefix"]
(ReqArg
(\arg opt -> return opt { optIdentifierPrefix = arg })
"STRING")
"" -- "Prefix to add to automatically generated HTML identifiers"
, Option "" ["indented-code-classes"]
(ReqArg
(\arg opt -> return opt { optIndentedCodeClasses = words $
map (\c -> if c == ',' then ' ' else c) arg })
"STRING")
"" -- "Classes (whitespace- or comma-separated) to use for indented code-blocks"
, Option "" ["toc", "table-of-contents"]
(NoArg
(\opt -> return opt { optTableOfContents = True }))
"" -- "Include table of contents"
, Option "" ["base-header-level"]
(ReqArg
(\arg opt ->
case reads arg of
[(t,"")] | t > 0 -> do
let oldTransforms = optTransforms opt
let shift = t - 1
return opt{ optTransforms =
headerShift shift : oldTransforms }
_ -> do
UTF8.hPutStrLn stderr $
"base-header-level must be a number > 0"
exitWith $ ExitFailure 19)
"NUMBER")
"" -- "Headers base level"
, Option "" ["template"]
(ReqArg
(\arg opt -> do
return opt{ optTemplate = Just arg,
optStandalone = True })
"FILENAME")
"" -- "Use custom template"
, Option "V" ["variable"]
(ReqArg
(\arg opt ->
case break (`elem` ":=") arg of
(k,_:v) -> do
let newvars = optVariables opt ++ [(k,v)]
return opt{ optVariables = newvars }
_ -> do
UTF8.hPutStrLn stderr $ "Could not parse `" ++ arg ++ "' as a key/value pair (k=v or k:v)"
exitWith $ ExitFailure 17)
"KEY:VALUE")
"" -- "Use custom template"
, Option "c" ["css"]
(ReqArg
(\arg opt -> do
-- add new link to end, so it is included in proper order
let newvars = optVariables opt ++ [("css",arg)]
return opt { optVariables = newvars,
optStandalone = True })
"URL")
"" -- "Link to CSS style sheet"
, Option "H" ["include-in-header"]
(ReqArg
(\arg opt -> do
text <- UTF8.readFile arg
-- add new ones to end, so they're included in order specified
let newvars = optVariables opt ++ [("header-includes",text)]
return opt { optVariables = newvars,
optStandalone = True })
"FILENAME")
"" -- "File to include at end of header (implies -s)"
, Option "B" ["include-before-body"]
(ReqArg
(\arg opt -> do
text <- UTF8.readFile arg
-- add new ones to end, so they're included in order specified
let newvars = optVariables opt ++ [("include-before",text)]
return opt { optVariables = newvars,
optStandalone = True })
"FILENAME")
"" -- "File to include before document body"
, Option "A" ["include-after-body"]
(ReqArg
(\arg opt -> do
text <- UTF8.readFile arg
-- add new ones to end, so they're included in order specified
let newvars = optVariables opt ++ [("include-after",text)]
return opt { optVariables = newvars,
optStandalone = True })
"FILENAME")
"" -- "File to include after document body"
, Option "T" ["title-prefix"]
(ReqArg
(\arg opt -> do
let newvars = ("title-prefix", arg) : optVariables opt
return opt { optVariables = newvars,
optStandalone = True })
"STRING")
"" -- "String to prefix to HTML window title"
, Option "" ["reference-odt"]
(ReqArg
(\arg opt -> do
return opt { optReferenceODT = Just arg })
"FILENAME")
"" -- "Path of custom reference.odt"
, Option "" ["epub-stylesheet"]
(ReqArg
(\arg opt -> do
text <- UTF8.readFile arg
return opt { optEPUBStylesheet = Just text })
"FILENAME")
"" -- "Path of epub.css"
, Option "" ["epub-cover-image"]
(ReqArg
(\arg opt ->
return opt { optVariables =
("epub-cover-image", arg) : optVariables opt })
"FILENAME")
"" -- "Path of epub cover image"
, Option "" ["epub-metadata"]
(ReqArg
(\arg opt -> do
text <- UTF8.readFile arg
return opt { optEPUBMetadata = text })
"FILENAME")
"" -- "Path of epub metadata file"
, Option "D" ["print-default-template"]
(ReqArg
(\arg _ -> do
templ <- getDefaultTemplate Nothing arg
case templ of
Right t -> UTF8.hPutStr stdout t
Left e -> error $ show e
exitWith ExitSuccess)
"FORMAT")
"" -- "Print default template for FORMAT"
2011-01-16 20:08:20 +01:00
, Option "" ["bibliography"]
(ReqArg
(\arg opt -> return opt { optBibliography = (optBibliography opt) ++ [arg] })
"FILENAME")
""
2011-01-16 20:08:20 +01:00
, Option "" ["csl"]
(ReqArg
(\arg opt -> return opt { optCslFile = arg })
"FILENAME")
""
2011-01-16 20:08:20 +01:00
, Option "" ["natbib"]
(NoArg
(\opt -> return opt { optCiteMethod = Natbib }))
"" -- "Use natbib cite commands in LaTeX output"
2011-01-16 20:08:20 +01:00
, Option "" ["biblatex"]
(NoArg
(\opt -> return opt { optCiteMethod = Biblatex }))
"" -- "Use biblatex cite commands in LaTeX output"
2011-01-16 20:08:20 +01:00
, Option "" ["data-dir"]
(ReqArg
(\arg opt -> return opt { optDataDir = Just arg })
"DIRECTORY") -- "Directory containing pandoc data files."
""
, Option "" ["dump-args"]
(NoArg
(\opt -> return opt { optDumpArgs = True }))
"" -- "Print output filename and arguments to stdout."
, Option "" ["ignore-args"]
(NoArg
(\opt -> return opt { optIgnoreArgs = True }))
"" -- "Ignore command-line arguments."
, Option "v" ["version"]
(NoArg
(\_ -> do
prg <- getProgName
UTF8.hPutStrLn stdout (prg ++ " " ++ pandocVersion ++ compileInfo ++
copyrightMessage)
exitWith ExitSuccess ))
"" -- "Print version"
, Option "h" ["help"]
(NoArg
(\_ -> do
prg <- getProgName
UTF8.hPutStr stdout (usageMessage prg options)
exitWith ExitSuccess ))
"" -- "Show help"
]
-- Returns usage message
usageMessage :: String -> [OptDescr (Opt -> IO Opt)] -> String
usageMessage programName = usageInfo
(programName ++ " [OPTIONS] [FILES]" ++ "\nInput formats: " ++
(intercalate ", " $ map fst readers) ++ "\nOutput formats: " ++
(intercalate ", " $ map fst writers ++ ["odt","epub"]) ++ "\nOptions:")
-- Determine default reader based on source file extensions
defaultReaderName :: String -> [FilePath] -> String
defaultReaderName fallback [] = fallback
defaultReaderName fallback (x:xs) =
case takeExtension (map toLower x) of
".xhtml" -> "html"
".html" -> "html"
".htm" -> "html"
".tex" -> "latex"
".latex" -> "latex"
".ltx" -> "latex"
".rst" -> "rst"
".lhs" -> "markdown+lhs"
".textile" -> "textile"
".native" -> "native"
".json" -> "json"
_ -> defaultReaderName fallback xs
-- Returns True if extension of first source is .lhs
lhsExtension :: [FilePath] -> Bool
lhsExtension (x:_) = takeExtension x == ".lhs"
lhsExtension _ = False
-- Determine default writer based on output file extension
defaultWriterName :: FilePath -> String
defaultWriterName "-" = "html" -- no output file
defaultWriterName x =
case takeExtension (map toLower x) of
"" -> "markdown" -- empty extension
".tex" -> "latex"
".latex" -> "latex"
".ltx" -> "latex"
".context" -> "context"
".ctx" -> "context"
".rtf" -> "rtf"
".rst" -> "rst"
".s5" -> "s5"
".native" -> "native"
".json" -> "json"
".txt" -> "markdown"
".text" -> "markdown"
".md" -> "markdown"
".markdown" -> "markdown"
".textile" -> "textile"
".lhs" -> "markdown+lhs"
".texi" -> "texinfo"
".texinfo" -> "texinfo"
".db" -> "docbook"
".odt" -> "odt"
".epub" -> "epub"
".org" -> "org"
['.',y] | y `elem` ['1'..'9'] -> "man"
_ -> "html"
main :: IO ()
main = do
rawArgs <- liftM (map decodeString) getArgs
prg <- getProgName
let compatMode = (prg == "hsmarkdown")
let (actions, args, errors) = if compatMode
then ([], rawArgs, [])
else getOpt Permute options rawArgs
unless (null errors) $
do name <- getProgName
mapM_ (\e -> UTF8.hPutStr stderr (name ++ ": ") >> UTF8.hPutStr stderr e) errors
UTF8.hPutStrLn stderr $ "Try " ++ name ++ " --help for more information."
exitWith $ ExitFailure 2
let defaultOpts' = if compatMode
then defaultOpts { optReader = "markdown"
, optWriter = "html"
, optStrict = True }
else defaultOpts
-- thread option data structure through all supplied option actions
opts <- foldl (>>=) (return defaultOpts') actions
let Opt { optTabStop = tabStop
, optPreserveTabs = preserveTabs
, optStandalone = standalone
, optReader = readerName
, optWriter = writerName
, optParseRaw = parseRaw
, optVariables = variables
, optTableOfContents = toc
, optTransforms = transforms
, optTemplate = templatePath
, optOutputFile = outputFile
, optNumberSections = numberSections
, optSectionDivs = sectionDivs
, optIncremental = incremental
, optOffline = offline
, optSmart = smart
, optHtml5 = html5
, optChapters = chapters
, optHTMLMathMethod = mathMethod
, optReferenceODT = referenceODT
, optEPUBStylesheet = epubStylesheet
, optEPUBMetadata = epubMetadata
, optDumpArgs = dumpArgs
, optIgnoreArgs = ignoreArgs
, optStrict = strict
Extensive changes stemming from a rethinking of the Pandoc data structure. Key and Note blocks have been removed. Link and image URLs are now stored directly in Link and Image inlines, and note blocks are stored in Note inlines. This requires changes in both parsers and writers. Markdown and RST parsers need to extract data from key and note blocks and insert them into the relevant inline elements. Other parsers can be simplified, since there is no longer any need to construct separate key and note blocks. Markdown, RST, and HTML writers need to construct lists of notes; Markdown and RST writers need to construct lists of link references (when the --reference-links option is specified); and the RST writer needs to construct a list of image substitution references. All writers have been rewritten to use the State monad when state is required. This rewrite yields a small speed boost and considerably cleaner code. * Text/Pandoc/Definition.hs: + blocks: removed Key and Note + inlines: removed NoteRef, added Note + modified Target: there is no longer a 'Ref' target; all targets are explicit URL, title pairs * Text/Pandoc/Shared.hs: + Added 'Reference', 'isNoteBlock', 'isKeyBlock', 'isLineClump', used in some of the readers. + Removed 'generateReference', 'keyTable', 'replaceReferenceLinks', 'replaceRefLinksBlockList', along with some auxiliary functions used only by them. These are no longer needed, since reference links are resolved in the Markdown and RST readers. + Moved 'inTags', 'selfClosingTag', 'inTagsSimple', and 'inTagsIndented' to the Docbook writer, since that is now the only module that uses them. + Changed name of 'escapeSGMLString' to 'escapeStringForXML' + Added KeyTable and NoteTable types + Removed fields from ParserState; 'stateKeyBlocks', 'stateKeysUsed', 'stateNoteBlocks', 'stateNoteIdentifiers', 'stateInlineLinks'. Added 'stateKeys' and 'stateNotes'. + Added clause for Note to 'prettyBlock'. + Added 'writerNotes', 'writerReferenceLinks' fields to WriterOptions. * Text/Pandoc/Entities.hs: Renamed 'escapeSGMLChar' and 'escapeSGMLString' to 'escapeCharForXML' and 'escapeStringForXML' * Text/ParserCombinators/Pandoc.hs: Added lineClump parser: parses a raw line block up to and including following blank lines. * Main.hs: Replaced --inline-links with --reference-links. * README: + Documented --reference-links and removed description of --inline-links. + Added note that footnotes may occur anywhere in the document, but must be at the outer level, not embedded in block elements. * man/man1/pandoc.1, man/man1/html2markdown.1: Removed --inline-links option, added --reference-links option * Markdown and RST readers: + Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, all note and key blocks are parsed on a first pass through the document. Once tables of notes and keys have been constructed, the remaining parts of the document are reassembled and parsed. + Refactored link parsers. * LaTeX and HTML readers: Rewrote to fit new Pandoc definition. Since there are no longer Note or Key blocks, notes and references can be parsed in a single pass through the document. * RST, Markdown, and HTML writers: Rewrote using state monad new Pandoc and definition. State is used to hold lists of references footnotes to and be printed at the end of the document. * RTF and LaTeX writers: Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Docbook writer: + Moved the functions 'attributeList', 'inTags', 'selfClosingTag', 'inTagsSimple', 'inTagsIndented' from Text/Pandoc/Shared, since they are now used only by the Docbook writer. + Rewrote using new Pandoc definition. (Because of the different treatment of footnotes, the "notes" parameter is no longer needed in the block and inline conversion functions.) * Updated test suite * Throughout: old haskell98 module names replaced by hierarchical module names, e.g. List by Data.List. * debian/control: Include libghc6-xhtml-dev instead of libghc6-html-dev in "Build-Depends." * cabalize: + Remove haskell98 from BASE_DEPENDS (since now the new hierarchical module names are being used throughout) + Added mtl to BASE_DEPENDS (needed for state monad) + Removed html from GHC66_DEPENDS (not needed since xhtml is now used) git-svn-id: https://pandoc.googlecode.com/svn/trunk@580 788f1e2b-df1e-0410-8736-df70ead52e1b
2007-04-10 03:56:50 +02:00
, optReferenceLinks = referenceLinks
, optWrapText = wrap
, optColumns = columns
, optEmailObfuscation = obfuscationMethod
, optIdentifierPrefix = idPrefix
, optIndentedCodeClasses = codeBlockClasses
, optDataDir = mbDataDir
, optBibliography = reffiles
, optCslFile = cslfile
, optCiteMethod = citeMethod
, optListings = listings
, optAscii = ascii
} = opts
when dumpArgs $
do UTF8.hPutStrLn stdout outputFile
mapM_ (\arg -> UTF8.hPutStrLn stdout arg) args
exitWith ExitSuccess
let sources = if ignoreArgs then [] else args
datadir <- case mbDataDir of
Nothing -> catch
(liftM Just $ getAppUserDataDirectory "pandoc")
(const $ return Nothing)
Just _ -> return mbDataDir
-- assign reader and writer based on options and filenames
let readerName' = if null readerName
then let fallback = if any isURI sources
then "html"
else "markdown"
in defaultReaderName fallback sources
else readerName
let writerName' = if null writerName
then defaultWriterName outputFile
else writerName
reader <- case (lookup readerName' readers) of
Just r -> return r
Nothing -> error ("Unknown reader: " ++ readerName')
templ <- case templatePath of
Nothing -> do
deftemp <- getDefaultTemplate datadir writerName'
case deftemp of
Left e -> throwIO e
Right t -> return t
Just tp -> catch (UTF8.readFile tp)
(\e -> if isDoesNotExistError e
then catch
(readDataFile datadir tp)
(\_ -> throwIO e)
else throwIO e)
let standalone' = standalone || isNonTextOutput writerName'
variables' <- case (writerName', standalone', offline) of
("s5", True, True) -> do
inc <- s5HeaderIncludes datadir
return $ ("s5includes", inc) : variables
("slidy", True, True) -> do
slidyJs <- readDataFile datadir $
"slidy" </> "slidy.min.js"
slidyCss <- readDataFile datadir $
"slidy" </> "slidy.min.css"
return $ ("slidy-js", slidyJs) :
("slidy-css", slidyCss) : variables
_ -> return variables
variables'' <- case mathMethod of
LaTeXMathML Nothing -> do
s <- readDataFile datadir $ "data" </> "LaTeXMathML.js"
return $ ("mathml-script", s) : variables'
MathML Nothing -> do
s <- readDataFile datadir $ "data"</>"MathMLinHTML.js"
return $ ("mathml-script", s) : variables'
_ -> return variables'
refs <- mapM (\f -> catch (readBiblioFile f) $ \e -> do
UTF8.hPutStrLn stderr $ "Error reading bibliography `" ++ f ++ "'"
UTF8.hPutStrLn stderr $ show e
exitWith (ExitFailure 23)) reffiles >>= \rs -> return $ concat rs
let sourceDir = if null sources
then "."
else takeDirectory (head sources)
let slideVariant = case writerName' of
"s5" -> S5Slides
"slidy" -> SlidySlides
_ -> NoSlides
let startParserState =
defaultParserState { stateParseRaw = parseRaw,
stateTabStop = tabStop,
stateLiterateHaskell = "+lhs" `isSuffixOf` readerName' ||
lhsExtension sources,
stateStandalone = standalone',
stateCitations = map refId refs,
stateSmart = smart || writerName' `elem`
["latex", "context", "latex+lhs", "man"],
stateColumns = columns,
stateStrict = strict,
stateIndentedCodeClasses = codeBlockClasses,
stateApplyMacros = writerName' `notElem` ["latex", "latex+lhs"] }
let writerOptions = defaultWriterOptions
{ writerStandalone = standalone',
writerTemplate = templ,
writerVariables = variables'',
writerEPUBMetadata = epubMetadata,
writerTabStop = tabStop,
writerTableOfContents = toc &&
writerName' /= "s5",
writerHTMLMathMethod = mathMethod,
writerSlideVariant = slideVariant,
writerIncremental = incremental,
writerCiteMethod = citeMethod,
writerBiblioFiles = reffiles,
writerIgnoreNotes = False,
writerNumberSections = numberSections,
writerSectionDivs = sectionDivs,
writerStrictMarkdown = strict,
writerReferenceLinks = referenceLinks,
writerWrapText = wrap,
writerColumns = columns,
writerLiterateHaskell = "+lhs" `isSuffixOf` writerName' ||
lhsExtension [outputFile],
writerEmailObfuscation = if strict
then ReferenceObfuscation
else obfuscationMethod,
writerIdentifierPrefix = idPrefix,
writerSourceDirectory = sourceDir,
writerUserDataDir = datadir,
writerHtml5 = html5 &&
"html" `isPrefixOf` writerName',
writerChapters = chapters,
writerListings = listings,
writerAscii = ascii }
when (isNonTextOutput writerName' && outputFile == "-") $
do UTF8.hPutStrLn stderr ("Error: Cannot write " ++ writerName ++ " output to stdout.\n" ++
"Specify an output file using the -o option.")
exitWith $ ExitFailure 5
let readSources [] = mapM readSource ["-"]
readSources srcs = mapM readSource srcs
readSource "-" = UTF8.getContents
readSource src = case parseURI src of
Just u | uriScheme u `elem` ["http:","https:"] ->
readURI u
_ -> UTF8.readFile src
readURI uri = simpleHTTP (mkRequest GET uri) >>= getResponseBody >>=
return . toString -- treat all as UTF8
let convertTabs = tabFilter (if preserveTabs then 0 else tabStop)
doc <- fmap (reader startParserState . convertTabs . intercalate "\n") (readSources sources)
let doc0 = foldr ($) doc transforms
doc1 <- if writerName' == "rtf"
then bottomUpM rtfEmbedImage doc0
else return doc0
doc2 <- do
if citeMethod == Citeproc && not (null refs)
then do
csldir <- getAppUserDataDirectory "csl"
cslfile' <- if null cslfile
then findDataFile datadir "default.csl"
else do
ex <- doesFileExist cslfile
if ex
then return cslfile
else findDataFile datadir $
replaceDirectory
(replaceExtension cslfile "csl")
csldir
processBiblio cslfile' refs doc1
else return doc1
case lookup writerName' writers of
Nothing | writerName' == "epub" ->
writeEPUB epubStylesheet writerOptions doc2
>>= B.writeFile (encodeString outputFile)
Nothing | writerName' == "odt" ->
writeODT referenceODT writerOptions doc2
>>= B.writeFile (encodeString outputFile)
Just r -> writerFn outputFile result
where writerFn "-" = UTF8.putStr
writerFn f = UTF8.writeFile f
result = r writerOptions doc2 ++
['\n' | not standalone']
Nothing -> error $ "Unknown writer: " ++ writerName'