The problem I am having with the LLVM-Haskell bindings is that I am getting "duplicated" names. I think the best way to explain my problem is with a small concrete example (note the example is contrived, and for such a small example there are easy ways around it... however it does point out my problem).
putc :: TFunction (Int32 -> IO Word32)
putc = newNamedFunction ExternalLinkage "putchar"
simple :: TFunction (Int32 -> IO Word32)
simple = do
internalputc <- putc
createNamedFunction ExternalLinkage "simple" $ \x -> do
call internalputc x
call internalputc x
ret (0 :: Word32)
easy :: TFunction (Int32 -> IO Word32)
easy = do
internalputc <- putc
internalsimple <- simple
createNamedFunction ExternalLinkage "easy" $ \x -> do
call internalsimple x
y <- add x (42 :: Int32)
call internalputc y
ret (0 :: Word32)
main :: IO ()
main = do
m <- newNamedModule "Main"
defineModule m easy
writeBitcodeToFile "SillyLib" m
If you now run this haskell program (you'll need some imports like Data.Int/Word, and LLVM.Core), you'll get the following output.
; ModuleID = 'SillyLib'
declare i32 @putchar(i32)
declare i32 @putchar1(i32)
define i32 @simple(i32) {
_L1:
%1 = call i32 @putchar1(i32 %0)
%2 = call i32 @putchar1(i32 %0)
ret i32 0
}
define i32 @easy(i32) {
_L1:
%1 = call i32 @simple(i32 %0)
%2 = add i32 %0, 42
%3 = call i32 @putchar(i32 %2)
ret i32 0
}
The problem is that in the IR, the (external) putchar is declared twice, but the second time with the name putchar1. I have a good sense as to why this is, but not a good sense for a nice general way around this. I.e. I don't want to have to put everything inside of one giant CodeGenModule.
This brings me to the another related issue. Is the LLVM-Haskell binding appropriate for building the backend of a compiler. Perhaps with a reasonable solution to the above -- I can figure out a way to use it ... but it seems simpler just to hand write the IR code...