Custom Query (964 matches)

In some computers, a segmentation violation occurred when reading the river vertically integrated mass transport (QbarG) in get_data.F:

     IF (LuvSrc(ng).or.LwSrc(ng)) THEN
       CALL get_ngfld (ng, iNLM, idRtra, SSF(ng)%ncid,                 &
    &                  1, SSF(ng), update(1),                          &
    &                  1, Nsrc(ng), 1, 2, 1, Nsrc(ng), 1,              &
    &                  SOURCES(ng) % QbarG(:,1))
       IF (exit_flag.ne.NoError) RETURN
     END IF

This was changed to remove (:,1) in QbarG:

     IF (LuvSrc(ng).or.LwSrc(ng)) THEN
       CALL get_ngfld (ng, iNLM, idRtra, SSF(ng)%ncid,                 &
    &                  1, SSF(ng), update(1),                          &
    &                  1, Nsrc(ng), 1, 2, 1, Nsrc(ng), 1,              &
    &                  SOURCES(ng) % QbarG)
       IF (exit_flag.ne.NoError) RETURN
     END IF

Similar changes are made to tl_get_data.F, rp_get_data.F and ad_get_data.F.

Many thanks to Teruhisa Okada for ​bringing this to my attention.

Some users are having issues with the FRCNAME global attribute in output NetCDF files because the list of files can be longer than the allocated length size for the string variable in def_info.F. This is solved by changing the length from 1024 to 1048.

      character (len=4096) :: string

Many thanks to Dezhou Yang, Kate Hedstrom and Kaitlin Alexander for ​reporting this issue.

Also, updated mod_scalars.F to initialize the logical switches LdefQCK(ng) and LwrtQCK(ng) asociated the the Quicksave output NetCDF file. This caused random crash behavior when running with lots of processors if the Quicksave output is off. This problem is typical with un-initialized variables and multiple processors. Such logical variables should be initialized to .FALSE. by the compiler, but sometimes it does not. It is always a good strategy to initialize all the variables in a Fortran program! I missed those.

Many thanks to Kaitlin Alexander for ​reporting this problem.

A factor of dt(ngc)/dt(ngf) was missing in the computation of the vertically integrated tracer flux correction (TFF) in two-way nesting in private routine correct_tracer_tile of nesting.F:

           TFF=TFF*dt(ngc)/dt(ngf)

The correction is small but it may make a difference in some applications at the refined grid boundary. It assumes that the coarse grid tracer flux remains constant over all fine grid mini-time steps.

Many thanks to Richard Leighton for his excellent ​description and solution to this problem. Check his ​application before and after the fix.