fix a bug found by Yosef Zlochower:

   In a multiprocessor run, AHFinderDirect finds different horizons
   in parallel on different processors, and doesn't broadcast horizon
   shapes to all processors unless it thinks they'll be needed.  However,
   the "the information will be needed" calculation doesn't take into
   account the possibility of the  HorizonRadiusInDirection()  aliased
   function being called on another processor, and in this case this
   function returns a wrong result if AHFinderDirect hasn't broadcast
   the horizon shapes.

* introduce a new Boolean parameter   always_broadcast_horizon_shape
  (defaults to false)
* document in interface.ccl, param.ccl, and doc/documentation.tex,
  that this parameter should be set to  true  if you expect
  HorizonRadiusInDirection() to be used
* inclusive-or this parameter into the "should we broadcast the horizon shape?"
  calculation in  src/driver/find_horizons.cc


git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinAnalysis/AHFinderDirect/trunk@1419 f88db872-0e4f-0410-b76b-b9085cfa78c5

1 files changed, 2 insertions, 1 deletions

diff --git a/src/driver/find_horizons.cc b/src/driver/find_horizons.cc
index e252528..a4d2b65 100644
--- a/src/driver/find_horizons.cc
+++ b/src/driver/find_horizons.cc
@@ -105,7 +105,8 @@ const int N_horizons = state.N_horizons;
 horizon_sequence& hs = *state.my_hs;
 const bool active_flag = hs.has_genuine_horizons();
 const bool broadcast_horizon_shape
-	= state.mask_info.set_mask_for_any_horizon
+	= state.always_broadcast_horizon_shape
+	  || state.mask_info.set_mask_for_any_horizon
 	  || state.store_info_in_SS_vars_for_any_horizon;
 
       struct cactus_grid_info&          cgi = state.cgi;