# Productization of ICON blueline

## Definition of production quality

???

## Support

Who is L1, L2, L3 for the different projects?

## Requirements

### Obvious Requirements

- Performance goal achieved
- Combine programs
- Improve GT4Py and DaCe transformations
- Functionality
- `concat_where`

### Requirements for minimally stable release (from maintenance perspective)

#### ICON-exclaim

- granule integration in the ICON build process
- delete Liskov
- memory consumption?

#### ICON4Py

- KHalfDim (needs Staggering)
- use field_operators instead of programs
- Verified on (in CI)
- Embedded
- target performance backend (DaCe)
- CI: Performance history
- Cleanup combined programs (?)
- init_with_zero

#### GT4Py

- remove `offset_provider`
- Staggering
- CI: Perfomance history
- Update to Python 3.12, 3.13 (for performance of Python parts)
- Generic Types for mixed precision cleanup (?)
- If performance without mixed presicion is sufficient, not needed for first release
- Expose `gpu`, `cpu` pointing to a performance backend.

### Other projects

- [x] GHEX publish source package to PyPi