{-# LANGUAGE CPP #-} {-# LANGUAGE OverloadedStrings #-} {-# LANGUAGE TemplateHaskell #-} module Data.FileEmbed.Inject ( -- * Inject into an executable -- $inject dummySpace , dummySpaceWith , inject , injectFile , injectWith , injectFileWith ) where import Language.Haskell.TH.Syntax (Exp (LitE), Lit (StringPrimL), Q) import qualified Data.ByteString as B import qualified Data.ByteString.Char8 as B8 import Data.ByteString.Unsafe (unsafePackAddressLen) import System.IO.Unsafe (unsafePerformIO) import Prelude as P magic :: B.ByteString -> B.ByteString magic x = B8.concat ["fe", x] sizeLen :: Int sizeLen = 20 getInner :: B.ByteString -> B.ByteString getInner b = let (sizeBS, rest) = B.splitAt sizeLen b in case reads $ B8.unpack sizeBS of (i, _):_ -> B.take i rest [] -> error "Data.FileEmbed (getInner): Your dummy space has been corrupted." padSize :: Int -> String padSize i = let s = show i in replicate (sizeLen - length s) '0' ++ s -- | Allocate the given number of bytes in the generate executable. That space -- can be filled up with the 'inject' and 'injectFile' functions. dummySpace :: Int -> Q Exp dummySpace = dummySpaceWith "MS" -- | Like 'dummySpace', but takes a postfix for the magic string. In -- order for this to work, the same postfix must be used by 'inject' / -- 'injectFile'. This allows an executable to have multiple -- 'ByteString's injected into it, without encountering collisions. -- -- Since 0.0.8 dummySpaceWith :: B.ByteString -> Int -> Q Exp dummySpaceWith postfix space = do let size = padSize space magic' = magic postfix start = B8.unpack magic' ++ size magicLen = B8.length magic' len = magicLen + sizeLen + space chars = LitE $ StringPrimL $ #if MIN_VERSION_template_haskell(2, 6, 0) map (toEnum . fromEnum) $ #endif start ++ replicate space '0' [| getInner (B.drop magicLen (unsafePerformIO (unsafePackAddressLen len $(return chars)))) |] -- | Inject some raw data inside a @ByteString@ containing empty, dummy space -- (allocated with @dummySpace@). Typically, the original @ByteString@ is an -- executable read from the filesystem. inject :: B.ByteString -- ^ bs to inject -> B.ByteString -- ^ original BS containing dummy -> Maybe B.ByteString -- ^ new BS, or Nothing if there is insufficient dummy space inject = injectWith "MS" -- | Like 'inject', but takes a postfix for the magic string. -- -- Since 0.0.8 injectWith :: B.ByteString -- ^ postfix of magic string -> B.ByteString -- ^ bs to inject -> B.ByteString -- ^ original BS containing dummy -> Maybe B.ByteString -- ^ new BS, or Nothing if there is insufficient dummy space injectWith postfix toInj orig = if toInjL > size then Nothing else Just $ B.concat [before, magic', B8.pack $ padSize toInjL, toInj, B8.pack $ replicate (size - toInjL) '0', after] where magic' = magic postfix toInjL = B.length toInj (before, rest) = B.breakSubstring magic' orig (sizeBS, rest') = B.splitAt sizeLen $ B.drop (B8.length magic') rest size = case reads $ B8.unpack sizeBS of (i, _):_ -> i [] -> error $ "Data.FileEmbed (inject): Your dummy space has been corrupted. Size is: " ++ show sizeBS after = B.drop size rest' -- | Same as 'inject', but instead of performing the injecting in memory, read -- the contents from the filesystem and write back to a different file on the -- filesystem. injectFile :: B.ByteString -- ^ bs to inject -> FilePath -- ^ template file -> FilePath -- ^ output file -> IO () injectFile = injectFileWith "MS" -- | Like 'injectFile', but takes a postfix for the magic string. -- -- Since 0.0.8 injectFileWith :: B.ByteString -- ^ postfix of magic string -> B.ByteString -- ^ bs to inject -> FilePath -- ^ template file -> FilePath -- ^ output file -> IO () injectFileWith postfix inj srcFP dstFP = do src <- B.readFile srcFP case injectWith postfix inj src of Nothing -> error "Insufficient dummy space" Just dst -> B.writeFile dstFP dst {- $inject The inject system allows arbitrary content to be embedded inside a Haskell executable, post compilation. Typically, file-embed allows you to read some contents from the file system at compile time and embed them inside your executable. Consider a case, instead, where you would want to embed these contents after compilation. Two real-world examples are: * You would like to embed a hash of the executable itself, for sanity checking in a network protocol. (Obviously the hash will change after you embed the hash.) * You want to create a self-contained web server that has a set of content, but will need to update the content on machines that do not have access to GHC. The typical workflow use: * Use 'dummySpace' or 'dummySpaceWith' to create some empty space in your executable * Use 'injectFile' or 'injectFileWith' from a separate utility to modify that executable to have the updated content. The reason for the @With@-variant of the functions is for cases where you wish to inject multiple different kinds of content, and therefore need control over the magic key. If you know for certain that there will only be one dummy space available, you can use the non-@With@ variants. -}