[cig-commits] [commit] devel: Update README_current_conclusions_from_Dimitri_about_the_bug.txt (08ae020)

[cig_noreply at geodynamics.org]

Repository : ssh://geoshell/specfem3d_globe

On branch  : devel
Link       : https://github.com/geodynamics/specfem3d_globe/compare/7ca3f38ee139abd7db6dff552b812d413432c105...08ae0200ffbe6dcfd0c61477d60ee11fc7ddf672

>---------------------------------------------------------------

commit 08ae0200ffbe6dcfd0c61477d60ee11fc7ddf672
Author: Dimitri Komatitsch <komatits at users.noreply.github.com>
Date:   Thu Apr 10 03:20:34 2014 +0200

    Update README_current_conclusions_from_Dimitri_about_the_bug.txt


>---------------------------------------------------------------

08ae0200ffbe6dcfd0c61477d60ee11fc7ddf672
 .../README_current_conclusions_from_Dimitri_about_the_bug.txt          | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/IMPORTANT_BUG_benchmark_Ebru_Par_file/README_current_conclusions_from_Dimitri_about_the_bug.txt b/IMPORTANT_BUG_benchmark_Ebru_Par_file/README_current_conclusions_from_Dimitri_about_the_bug.txt
index 97be74f..8f4c088 100644
--- a/IMPORTANT_BUG_benchmark_Ebru_Par_file/README_current_conclusions_from_Dimitri_about_the_bug.txt
+++ b/IMPORTANT_BUG_benchmark_Ebru_Par_file/README_current_conclusions_from_Dimitri_about_the_bug.txt
@@ -1,4 +1,3 @@
-
 The three input files needed: Par_file, CMTSOLUTION and STATIONS are in this directory. (and should be moved to directory "DATA" before starting a run)
 
 ====================
@@ -15,6 +14,8 @@ the bug does not come from option -ftz (Flush to Zero) used to compile the code,
 
 the bug is probably not in the MPI routines, since the 1D attenuation benchmark is OK (and the MPI routines do not know if the model is 1D or 3D), and last summer Daniel rewrote the MPI routines almost entirely but the bug is still there
 
+the bug is not something obvious such as an array index going out of bounds, because the code does not display any error message for that benchmark when running with full range checking options (as well as stack checking, unassigned variable detection and so on)
+
 Ebru's current runs are highly affected (the example is her Par_file, with a single change: 60 min instead of 100 min for the total duration of the seismograms, just to reduce the cost of the tests)
 
 The bug has been there for at least a year, probably more (my comment about this in the to-do list was from August 2013, back then I thought the switch to Daniel's new version at the end of Aug 2013 would fix it, but unfortunately it is still there); thus it is likely in a part of the code that did not change much when we switched to the new version