Skip to content

Avoid casting field input data to float32 #1607

@erikvansebille

Description

@erikvansebille

Early on in the design of Parcels (pre-v2), we decided to do most computations in float32. With the development of C-grid interpolation in v2, we required float64 accuracy to properly deal with boundary conditions. Still, we left a lot of float32 support under the hood in Parcels

With the recent release of NumPy v2, some of these float32-implementations seem to be breaking some of the unit tests. This probably has to do with this change to data type promotion in NumPy. See also #1603 (comment).

This has been quick-fixed in 20206c2, but ideally, we may want to reconsider the use of float32 in Parcels altogether.

This Issue is a place to discuss/explore what would happen if we move fully to float64 in Parcels

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    Status

    Done

    Status

    Done

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions