Fix logic around g2x and x2g vector index calculation #7505
Conversation
Calculate all g2x and x2g indices at the same time based on first_time logic, which is necessary when x2g and g2x calculations are split between calls to budgets1 and budgets2.
stephenprice
added
mpas-albany-landice
Coupler
There was a problem hiding this comment.
Approving based on inspection and previous discussion. The previous implementation meant that the first_time operations would only be done for one or the other coupling direction, which is obviously fragile. It looks like some other components do things the new way in this PR to avoid that, but atm, lnd, ocn, and ice do it the old way. They must be avoiding this error just be chance of how their seq_diag_*_mct routines are called.
@rljacob, cc'ing you so that's on your radar.
|
I noticed that sea ice also splits between budgets1 and budgets2, and they get around this issue by simply re-calculating indices every time in budgets1 (the first_time logic is only employed for budgets2). |
…xt (PR #7505) Fix logic around g2x and x2g vector index calculation This PR corrects the "first_time" logic around the calculation of indices for the glc coupling vectors g2x and x2g. Because the calls to x2g and g2x happen in budgets1 and budgets2, respectively, both sets of indices need to be calculated at the same time if the first_time logic is to be used so that these calculations only happen once. This fix has been tested and confirmed working. [BFB]
|
Passes:
merged to next |
|
merged to master |
Fix logic around g2x and x2g coupler vector index calculation.
This PR corrects the "first_time" logic around the calculation of indices for the glc
coupling vectors g2x and x2g. Because the calls to x2g and g2x happen in budgets1
and budgets2, respectively, both sets of indices need to be calculated at the same time
if the first_time logic is to be used so that these calculations only happen once. This fix
has been tested and confirmed working.