[project @ 2003-07-24 07:38:54 by simonpj]

[ghc-hetmet.git] / ghc / driver / ghci / ghci.c

diff --git a/ghc/driver/ghci/ghci.c b/ghc/driver/ghci/ghci.c
index 8c6926c..c0e3991 100644 (file)
--- a/ghc/driver/ghci/ghci.c
+++ b/ghc/driver/ghci/ghci.c
@@ -1,9 +1,13 @@
 /*
  *
- * $Id: ghci.c,v 1.6 2003/06/02 16:24:07 sof Exp $
+ * $Id: ghci.c,v 1.8 2003/06/12 09:48:17 simonpj Exp $
  *
- * ghci wrapper - invokes ghc.exe with the added command-line
+ * ghci wrapper for Win32 only
+ * 
+ * This wrapper invokes ghc.exe with the added command-line
  *                option "--interactive".
+ * (On Unix this is done by the ghci.sh shell script, but
+ *  that does not work so well on Win32.)
  *
  * (c) The GHC Team 2001
  *
@@ -118,8 +122,25 @@ main(int argc, char** argv)
      Cygwin gives me the right behaviour, but does it by
      implementing it in terms of spawnv(), so you pay
      the cost of having to create an extra process.
+     Plus, of course, we aren't allowed to use Cygwin here, because
+     GHC does not assume Cygwin.
      
-     ==> Just use spawnv().
+     ==> Just use spawnv(), which is provided by msvcrt.dll, the
+         Microsoft C runtime to which mingw delegates almost all
+        system calls
+
+        [Sigbjorn adds 12 Jun 03]
+     We probably ought to use CreateProcess() in ghci.c -- or better still an exec()-like
+     that didn't have to create a separate process from the wrapper (which is what that
+     code comment in there is driving at.) 
+      
+     CreateProcess() is a more wieldy function to invoke, which is probably why
+     I opted for spawnv(). spawnv() performs the equivalent of Prelude.unwords
+     (to look at the code itself, or at least an older version, see dospawn.c in the
+     vc98/crt/src/ directory of an MSVC6 installation.)
+      
+     CreateProcess() is a native Win32 API though, which has the merit that it is
+     guaranteed to work the same with both the mingw and cygwin ports.
   */
 #if 0
   fprintf(stderr, "Invoking ghc: ");