A Haskell FFI calling convention for Javascript
2010/10/29 5 Comments
Haskell’s Foreign Function Interface (FFI) predefined calling conventions do not match well with Javascript’s object oriented features. In particular selecting a field of an object using dot notation (like o.f
) and using an object as an array using brackets (like o[i]
) do not have a natural counterpart in Haskell or the default calling conventions supported by the FFI interface. So, here are some examples of how Javascript is accessed in UHC via its jscript calling convention:
data Document foreign import jscript "document" document :: IO Document foreign import jscript "%1.write()" documentWrite :: Document -> JSString -> IO () foreign import jscript alert :: JSString -> IO ()From within a browser the document representation can be accessed via the global variable
document
, the foreign entity "document"
translates to a reference to this variable. The type of the document is defined as an opaque type, it can thus only manipulated via Javascript. Writing a string to the document is done by invoking the method write on a document. The foreign entity "%1.write()"
specifies that from all arguments the first one is used as the receiver of the method write. The parenthesis ()
specify that a call has to be made, passing all arguments except those referred to explicitly by means of %<nr>
, where <nr> >= 1
refers to argument <nr>. If an entity is omitted as in alert it defaults to "<functionname>()"
where <functionname>
is the name of the foreign function being defined.
Function documentWrite does not accept a String but a JSString instead, defined to be the platform dependent representation of Strings, converted to and from String with corresponding conversion functions.
type JSString = PackedString stringToJSString :: String -> JSString jsStringToString :: JSString -> StringstringToJSString forces its argument to be fully evaluated and then converts it to a Javascript String. There is choice whether to put
document
in the IO monad or not, depending whether this global object itself will ever be assigned a new value or not. Not being a Javascript DOM wizard wrapping in IO seems to be the safest bet.
Given these functions a minimal Hello World web program thus is:
main = alert $ stringToJSString "Hi there!"As this would pop up an alert box, an alternative Hi is the following program which writes to the document instead:
main = do d <- document documentWrite d $ stringToJSString "Hi there!"Actually, the usual Hello would have worked as well because it is implemented as writing to the document:
main = putStr "Hi there!"To show the usefulness of array like access as part of we do a bit of rudimentary DOM programming:
foreign import jscript "%1.getElementsByName()" documentGetElementsByName :: Document -> JSString -> IO (NodeList Node) data NodeList x foreign import jscript "%1.length" nodeListLength :: NodeList Node -> Int foreign import jscript "%1[%2]" nodeListItem :: NodeList Node -> Int -> IO Node data Node foreign import jscript "%1.innerHTML" elementInnerHTML :: Node -> JSString foreign import jscript "%1.tagName" elementTagName :: Node -> JSStringA NodeList is not an array, but behaves like an array: we can ask for its length and retrieve an element by index. It is not an array itself, so modelling it as such in Haskell would be incorrect. However, by allowing import entities to use Javascript array notation we circumvent this limitation and the Javascript array interface can still be used easily. Finally, this minimal interface to DOM can be used to retrieve and print info about an element in an html document:
main = do d <- document nl <- documentGetElementsByName d (stringToJSString "myHeader") print (nodeListLength nl) n <- nodeListItem nl 0 print $ jsStringToString $ elementTagName n print $ jsStringToString $ elementInnerHTML nGiven the presence of
with the name "myHeader" in the document where the program is run, it will produce the following as part of the document:Head says hello!
1 "H1" "Head says hello!"
Interesting!
On thing: NodeList is a live list, so I’d say that nodeListLength needs to be in IO as well.
DOM programming would seem to be modelled by STM. JavaScript interpreters in browsers (and many outside them) are single-threaded; so there is never any danger of concurrent update of a structure during the “check” phase right before a transaction to the DOM is committed.
That’s wrong – Chrome and Opera JS engines are multithreaded, although they lock DOM accesses to ensure compatibility with the legacy JS code.
So, it boils down to putting everything in the IO monad, as all fields of an object (except the readonly ones) can then be modified by another thread, no way around it. Which is somewhat of a pity as it does not improve readability.
You post very interesting posts here. Your website deserves much more visitors.
It can go viral if you give it initial boost, i know useful tool that can help you, just search
in google: svetsern traffic tips