Opened 6 months ago

Closed 4 months ago

#1988 closed Bug (fixed)

AGRIF code corruption test fails in SETTE

Reported by:Claire Levy Component: AGRIF
Version: release-3.6 Severity: minor
Keywords: Cc:
Management
Owned by: Milestone: 2017 WP
Priority: high

Description

Context

Within the sequence of SETTE tests, a test to check that AGRIF does not corrupt the whole code has been implemented 2 years ago (changeset 5588 for 3_6 release, and changeset 5589 for trunk.
It seems these tests were not taken in account in the sette report (sette_rpt.sh) up to now

Analysis

In new_sette.sh and new_sette_rpt.sh (in the trunk for now), this test is now indeed checked and indicates that the results are indeed different whereas they should not: the 2 solver.Stat are indetical for 39 time steps, and differ afterwards

Fix

Fix depends en detailed analysis to be done:

  • if is should be considered that it is OK to have different solver.stat after 39 time steps, result of this test should be changed in new_sette_rpt.sh
  • otherwise, it seems there is a bug in AGRIF implementation…, to be fixed in trunk and 3_6 release

Commit History (0)

(No commits)

Change History (3)

comment:1 Changed 6 months ago by Jerome Chanut

set ln_icebergs=.false. in sette tests (n°14), this should fix the problem (if ln_icebergs=T, Agrif set it to false anyway so that one simulation runs with icebergs and the one compiled with key_agrif no. The best thing to do, would to set ln_icebergs=F on child grids only).


comment:2 Changed 6 months ago by Claire Levy

  • Version changed from trunk to release-3.6

Yes indeed, setting ln_icebergs=.false. in sette tests does make the non corruption test successful. It is now fixed in new_sette.sh in the trunk and I leave the ticket open, moving it to 3_6_STABLE so that we dont forget to propagate this fix back into SETTE of 3_6_STABLE

comment:3 Changed 4 months ago by Jerome Chanut

  • Resolution set to fixed
  • Status changed from new to closed
Note: See TracTickets for help on using tickets.