Using the ghc-api to run more than one instance of ghc simultaneously

[Simon Marlow]

On 12/07/2009 16:49, Daniel Gorín wrote:
> Hi
>
> I'm trying to use the GHC API to have several instances of GHC's
> interpreter loaded simultaneously; each with its own loaded modules,
> etc. However, this doesn't seem to work well when two instances have
> loaded modules with the same name. I'm including the code of a
> small(ish) example of this at the end of the message.
>
> The example launches two threads (with forkIO) and fires GHC in
> interpreted mode on each thread (with GHC.runGhc); then it sequentially
> loads file TestMain1.hs in the first and TestMain2.hs in the second one
> and finally tries to evaluate expression test1 defined in the first one
> followed by test2 defined in the second one. The output is:
>
> #./Main
> 1: Load succeded
> 2: Load succeded
> 3: (1,2,3)
> 4: Main:
> During interactive linking, GHCi couldn't find the following symbol:
> Main_test1_closure
> This may be due to you not asking GHCi to load extra object files,
> archives or DLLs needed by your current session. Restart GHCi, specifying
> the missing library using the -L/path/to/object/dir and -lmissinglibname
> flags, or simply by naming the relevant files on the GHCi command line.
> Alternatively, this link failure might indicate a bug in GHCi.
> If you suspect the latter, please send a bug report to:
> glasgow-haskell-bugs at haskell.org
>
> Main: thread blocked indefinitely
> #
>
> The "thread blocked indefinitely" message is not important (comes from
> simplifying the original example). I tried this both in ghc 6.10.1 and
> ghc 6.11.20090607 with the same results.
>
> Is this a known limitation? Or should I be doing it some other way?

Yes, it is a known limitation.  It ought to be documented somewhere.

There are two problems:

  1. GHC is not thread-safe.  There are some items of global state (the
     NameCache and the PackageInterfaceTable) that should be protected.
     It's not a trivial matter to do this - we had a hacky solution for
     the parallel compilation experiment we did a while back, but it
     wasn't finished.

     You could workaround this by using a mutex and only invoking GHC
     API operations in one thread at a time.

  2. There is only one RTS linker with a single symbol table.  This is
     the problem you ran into.  There's no workaround that I'm aware
     of.

Cheers,
	Simon