Today's meeting did not have a specific agenda so we took turns giving updates.
Yannick opened by mentioning that we have two variants of OOPS develop branches (develop-bump and develop-bmatrix), and he would like to get these merged next week into the develop branch. Yannick will send out a message today that explains the status of these branches and the plan for merging them together.
Yannick noted that there still exists some minor issues with the Intel compiler in regard to these branches and the core team will be working on resolving these.
Xin announced that he is having issues compiling inside the Singularity container with GNU compilers.
- fv3-bundle
- Error message: No CommFactory for "serial"
- Chris H explained later on that this is coming from eckit and has to do with constructing an MPI communicator object
- This was working on Monday but on Wednesday this was broken (what has changed since then?)
- Ben Johnson said that he is seeing the same message when inside Signularity
- The core team will also look into resolving this issue.
NCAR
Chris said that they have been having some issues with the Intel compiler and wondered if these are related to the issues mentioned above. He will send out the error message they are seeing.
BJ got an example working for univariate 3DVar covariance with MPAS and BUMP. He needed to checkout or merge the OOPS feature/anisotropic_correlations branch to get this working.
JJ is working on getting mpas-bundle working on Cheyenne. He is seeing issues with the hofx test, and mentioned that the "mpi.h" file is not being found during the test. Chris H. responded that this is part of the fix he recently made for MPI and that they need to add the appropriate include statement to the routine that the failure is coming from.
There is general confusion about which OOPS branches should be used. Yannick's message mentioned above should help clarify.
College Park
Guillaume is doing some cleanup from the bmatrix code sprint. He is currently okay with working on his laptop, but needs the HPC system soon and is experiencing Intel compiler issues. Hopefully, this will be covered by the work that the core team is doing in regard to the compiler issues Yannick mentioned above.
Andrew reported that he is getting up to speed on JEDI and is eager to contribute. He requested help with QC for the Radiance obs type. Yannick said that a separate call can be setup soon to discuss how best to proceed. Andrew asked about the status of the Locations class, and Yannick responded that there have been some discussions in the GNSSRO code sprint this week and that we can continue discussions later on. Expect a briefing from the code sprint at next week's meeting.
Rahul is working on getting JEDI functioning on a NASA workstation. He is also working on converting NASA data tanks to ncdiag, and needs to have more discussions about the requirements for marine obs types with Guillaume.
UKMO
Steven asked for clarification on the OOPS develop branches. Yannick's message mentioned above should address this. Steven is working on LFRic integration and is running into issues with the OOPS develop variant branches. Yannick added to not spend effort with the develop-bump branch since develop-bmatrix is newer. Steven asked if develop-bump is working with MPAS, and BJ responded yes.
Marek mentioned that develop-bmatrix doesn't have the new MPI functionality (from Chris H) and this may be causing issues. Yannick responded that this was so because some other developers were not ready for the MPI changes. Yannick's message mentioned above should help clarify how people working with specific models in JEDI should proceed in regard to the OOPS develop branches.
The develop-bump branch of oops has the new MPI functionality. The develop-bmatrix branch does not, but it has some updates to the bump implementation that users may find helpful. We're not ready to merge these two branches yet because it may cause problems for some, but Yannick will create a new branch today that combines the up-to-date functionality of both.
Boulder
Steve H is working on installing the April 15, 2018 00Z geovals and obs data into JEDI. He has files ready for Radiosonde and AMSU-A, and will have Aircraft ready soon. The marine files are already built for the April 15th date. Steve is reorganizing the locations where these test files will reside. The geovals data will live in ufo/test/ and the obs data will live in ioda/test. Steve asked Guillaume what should be done with unused (at this point) marine files Jason-3-2018-04-15.nc and cryosat2-2018-04-15.nc. Ie, do they belong in ufo or ioda. Guillaume responded that these are under development so we can just leave them where they are for now.
Mark and Xin are updating the container with the latest versions of eckit and fckit. Using these requires a simple change to OOPS (one file). As part of this work, a bug was fixed in ecbuild. Merging this in can wait until next week when we deal with merging the OOPS develop variant branches.
BJ commented about a 4Dvar issue where all obs are being updating on every time step (instead of selectively updating the appropriate obs at each time step). Yannick acknowledged that this is a bug which will be fixed after the April 15th data is installed.