Updating static files never triggers recompilation

[snoyberg]

Consider the following code:

{-# LANGUAGE OverloadedStrings #-}

module Main where

import IdeSession
import Data.Monoid (mconcat)
import qualified Data.ByteString.Lazy.Char8 as L8

main :: IO ()
main = do
    sess <- initSession defaultSessionConfig
                { configDir             = "."
                }

    let cb = \_ -> return ()
        update = flip (updateSession sess) cb

    let mainContents = L8.pack $ unlines
            [ "{-# LANGUAGE TemplateHaskell #-}"
            , "import Text.Hamlet"
            , "import Text.Blaze.Html.Renderer.String"
            , "main = putStrLn $ renderHtml $(shamletFile \"foo.hamlet\")"
            ]

    update $ mconcat
        [ updateModule "Main.hs" mainContents
        , updateDataFile "foo.hamlet" "<p invalid"
        ]

    -- Error message expected, invalid data file
    getSourceErrors sess >>= print

    update $ updateDataFile "foo.hamlet" "<p>Valid"

    -- No error message expected, but got one anyway
    getSourceErrors sess >>= print

    update $ updateModule "Main.hs" $ mainContents `L8.append` "\n\n-- Trigger a recompile"
    -- No more error message
    getSourceErrors sess >>= print

    update $ updateDataFile "foo.hamlet" "<p invalid"

    -- Error message expected, but recompilation is not triggered.
    getSourceErrors sess >>= print

I've added an invalid Hamlet file, and a source file which references this Hamlet file. As expected, compilation fails. However, updating the data file with a valid Hamlet file does not trigger any recompilation, and the only way to get the change noticed is to actually change the source file. Going in the reverse direction, updating the Hamlet file with something invalid does not trigger recompilation either.

Note that Hamlet uses qAddDependentFile (http://hackage.haskell.org/packages/archive/template-haskell/2.7.0.0/doc/html/Language-Haskell-TH-Syntax.html#v:qAddDependentFile) internally, so under normal circumstances GHC is able to detect that a recompilation is necessary.

[snoyberg]

Note: this is not a high priority, urgent issue, but I would like to get some feedback on what's going on here and how difficult it would be to fix.

[edsko]

We keep a flag internally whether we should invoke ghc at the next call to updateSession, and this flag is set whenever you change a source file, but not when you change a data file. So the easy fix to this would be to set this flag on updateDataFile too. The better fix would be to hook in to qAddDependentFile, but that's probably overkill. A third option would be to add an extra argument to updateDataFile to specify whether this should trigger recompilation or not.

[snoyberg]

Let's go for the easy fix: just setting the flag on updateDataFile. If you can provide a commit on top of the temporary-pkgdir-hack branch, we can start testing this and see if it solves our problem (while avoiding any negative side-effects).

[Mikolaj]

Pinging @snoyberg and @jwiegley, just in case. :)

[snoyberg]

Perfect, thanks. And yes, the ping was very good, we don't get notifications when commits have been made.

@jwiegley Can you update our ide-backend on master to use this change?

[jwiegley]

@snoyberg Doing so now.

[snoyberg]

This is now on our staging system, and does indeed solve our problem. Thank you for the fast turnaround @edsko and @Mikolaj. Closing.

[Mikolaj]

Thank you.