Opened 5 years ago
Closed 5 years ago
#2085 closed Bug (fixed)
dev_merge_2017/trunk without land-porc crashes when starting from restart
| Reported by: | cbricaud | Owned by: | cbricaud |
|---|---|---|---|
| Priority: | low | Milestone: | |
| Component: | OCE | Version: | trunk |
| Severity: | minor | Keywords: | |
| Cc: |
Description (last modified by nicolasmartin)
Context
run without land-proc
start from restart produced by a previous run
=> crash when reading restart:
Analysis
iom.F90/iom_open":
llnoov = .NOT. lk_agrif
so restart files are written without overlapping for both cases with/without land-proc
iom.F90/iom_get_123d:
llnoov = (jpni * jpnj ) == jpnij .AND. .NOT. lk_agrif
so restart files are read without overlapping for case with land-proc
so restart files are read with overlapping for case without land-proc
Fix
iom.F90/iom_open":
replace
llnoov = .NOT. lk_agrif
by
llnoov = (jpni * jpnj ) == jpnij .AND. .NOT. lk_agrif
Commit History (2)
| Changeset | Author | Time | ChangeLog |
|---|---|---|---|
| 10016 | cbricaud | 2018-07-31T09:03:08+02:00 | commit bugfix for ticket #2085 |
| 10015 | davestorkey | 2018-07-30T14:16:24+02:00 | UKMO dev_r9950_GO8_package branch: include fix for restartability with land suppression as temporary measure before this is fixed in the NEMO trunk (NEMO ticket #2085). |
Change History (8)
comment:1 Changed 5 years ago by smasson
comment:2 Changed 5 years ago by cbricaud
I tried it
it crashes in mpp-ini
comment:3 Changed 5 years ago by nicolasmartin
- Description modified (diff)
comment:4 Changed 5 years ago by nemo
- Owner set to cbricaud
- Status changed from new to assigned
comment:5 Changed 5 years ago by mathiot
Tests on ORCA2_ICE_PISCES:
- I tried land suppression with the fix suggested by Clement and it works for me.
- I tried by setting llnoov to .TRUE. as Sebastien suggested (we only read the interior domain data), and it failed with NaN. To have it working, I need to call twice the lbc_lnk instance and the end of iom_get_123d. I think this is an issue with the initialisation of corner. One call to lbc_lnk is not enough to initialised correctly the corners of a domain south/north of a land processor as we send data E/W (corner not filled as no data over halo on the E or W processor) then N/S (corner not filled as the N or S processor is a land processor).
comment:6 Changed 5 years ago by davestorkey
In 10015:
comment:7 Changed 5 years ago by cbricaud
In 10016:
comment:8 Changed 5 years ago by cbricaud
- Resolution set to fixed
- Status changed from assigned to closed
Note: See
TracTickets for help on using
tickets.
Why not doing the opposite ?
in iom.F90/iom_get_123d:
replace
by
I don't really understand why we have to keep this llnoov. To me it should always be true