ert
ert copied to clipboard
equinor
Remove EclConfig
It seems like we might be able to remove ecl config altogether, after all the dust has settled.
At the time of writing, it holds three data:
- num cpu
- grid (and its file)
- refcase (and its file)
the files are only used to load grid and refcase. there is a PR ready for review right now which removes num cpu from ecl config. investigating uses of grid and refcase shows that, besides tests:
ecl config's grid is
- passed to ensemble config in res config's init,
- used to construct observations in enkfmain, and
- mentioned in a comment about row scaling (which i sadly don't really understand)
while ecl config's refcase is
- also used to construct observations in enkfmain, and
- attached to the time map in storage if defined, and
- used to extract data in libres facade
- passed on to both ensemble and model config in res config's init
my idea is
- to have the grid live only in ensemble config, and passed from there to construct observations in enkf main, and
- to have refcase reside in ensemble config~~res config~~, and passed down from there to ~~ensemble and~~ model config as a boolean, as per its use described in a comment below.
~~but for the second point, i shall investigate usage of refcase in both ensemble and model config, and see if any of them can actually do without, and then adjust the plan accordingly.~~
seems like ensemble config actually uses grid, to construct fields, based on values from the config option FIELD. and that one is used quite a bit, according to our logs from the last 3 months-ish.
it also uses refcase, to add "full" summaries for each element of the list given by the SUMMARY config option. and that config option is used a lot, again according to our logs.
so no dice there, these need to stay, ensemble config needs both grid and refcase.
model config just gets / uses refcase. here, things get interesting - model config doesn't really use the structure of refcase, it ever just assigns to the respective struct field, and checks if it's not NULL
so we could substitute a simple boolean here, something along the lines of refcase_given.
in the python code, there is a check if the refcase is an instance of EclSum - not sure if there are historical reasons for that check, but it seems to me like it is a security redundancy, that can be scrapped.
in conclusion - we should be able to let refcase live in ensemble config, and just pass a boolean to model config that reflects whether ensemble config holds a ref case.
updated idea point two accordingly.
actually as @andreas-el pointed out kindly, model config doesn't just use ref case as a pure exists or not boolean, but passes it on to some function residing in ecl connected to history / last report / something (ecl_sum_get_last_report_step)
maybe we can still save the above considerations, and just pass the result of ecl_sum_get_last_report_step(refcase) to model config!
reopening - we're only almost there, still need #4101 to be merged, which takes care of num cpu, then we can make a PR that actually just removes the no longer used ecl config.
reopening - we're only almost there, still need #4101 to be merged, which takes care of num cpu, then we can make a PR that actually just removes the no longer used ecl config.
Yes, this makes sense. We also need to do some cleanup after num_cpu is resolved. Adds PR for that too. #4131
