Closed bennybp closed 2 years ago
Doesn't this change only impact how the records are accessed? From the op it looks like the computation has been split among tasks, with errors occurring at compute time on the manager under psi4
.
The composition planner is called both for computing and accessing:
I guess there may still be a problem with existing records? But with the way datasets are currently written that shouldn't be an issue. The previously submitted ones will disappear from the dataset (because the dataset doesn't explicitly link to individual records)
The composition planner is called both for computing and accessing:
I guess there may still be a problem with existing records? But with the way datasets are currently written that shouldn't be an issue. The previously submitted ones will disappear from the dataset (because the dataset doesn't explicitly link to individual records)
Ah perfect, you're right, missed the usage in Dataset._compute
. With this change use of a new qcportal
from a new QCFractal release should generate the right set of records when called. Can basically try to resubmit the dataset and this should just work.
Description
Fixes issue #688
These functionals shouldn't be split in the composition planner, as the first part isn't available in psi4 as a plain functional.
This is a temporary fix. The next major version will probably do away with the composition planner altogether.
Changelog description
Skip composition planner for certain functionals
Status