refactor[next]: Use cartesian shift syntax in all tests

[tehrengruber]

Prerequisite for staggered fields (#2339), split out so that PR reduces to the staggering feature itself.

Changes

• New CartesianOffset IR node. field(Dim ± n) now lowers to shift(cartesian_offset(Dim, Dim), n) — a node carrying the offset's source/target dimensions with their DimensionKind — replacing the implicit string offset (_OffDim), which dropped the kind and broke vertical KDim ± n on the gtfn/dace/roundtrip pipelines. Handled in type inference/synthesis, domain inference, trace_shifts, pretty-print/parse, and the gtfn/dace/roundtrip backends (roundtrip cartesian-offset support was previously missing). The shift no longer needs an offset-provider entry — the node is self-describing.
• Test migration. All tests use the field(Dim ± n) syntax; the Ioff/Joff/Koff FieldOffsets and their offset-provider entries are removed, kept only where still required (dynamic as_offset, iterator-level string offsets).

Why the node carries domain and codomain

CartesianOffset encodes a source→target dimension mapping. In this PR every cartesian shift is a translation (domain == codomain, e.g. IDim → IDim), enforced by explicit assert domain == codomain in the domain-inference and backend handlers. The second dimension exists for #2339's relocation shifts (domain != codomain, e.g. IDim → IHalfDim for half-cell ±0.5 staggered shifts): keeping the node shape here lets #2339 add relocation by dropping those asserts and implementing the domain != codomain case — without reshaping the node or re-touching every handler.

Out of scope (stays in #2339)

No OffsetProviderElem retype (bare Dimension is still valid in offset providers), no Connectivity Protocol→concrete refactor, and no staggering (_Staggered dims / ±0.5 half shifts).

[Copilot]

Pull request overview

This PR introduces a dedicated CartesianOffset iterator-IR node to represent cartesian shifts (field(Dim ± n)) while preserving dimension kind (notably vertical KDim), and migrates the test suite to consistently use the cartesian shift syntax instead of Ioff/Joff/Koff FieldOffsets.

Changes:

• Add itir.CartesianOffset plus type inference/synthesis, domain inference/translation, pretty-printing, and backend support (gtfn, DaCe, roundtrip).
• Update ffront lowering so field(Dim ± n) lowers to shift(cartesian_offset(Dim, Dim), n) instead of implicit string offsets.
• Migrate tests to field(Dim ± n) and remove most Ioff/Joff/Koff usage and offset-provider entries where no longer needed.

Reviewed changes

Copilot reviewed 27 out of 27 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/next_tests/unit_tests/iterator_tests/test_type_inference.py	Drop Ioff/Koff fixture imports where tests now rely on cartesian shift encoding.
tests/next_tests/unit_tests/ffront_tests/test_foast_to_gtir.py	Update reference IR to expect CartesianOffset in lowered shift.
tests/next_tests/unit_tests/embedded_tests/test_basic_program.py	Replace FieldOffset-based shift with IDim + 1 and remove offset_provider from call site.
tests/next_tests/integration_tests/multi_feature_tests/ffront_tests/test_laplacian.py	Remove Joff import (tests now use cartesian shift syntax).
tests/next_tests/integration_tests/multi_feature_tests/ffront_tests/test_icon_like_scan.py	Replace Koff[...] shifts with KDim ± n; stop providing Koff in offset_provider.
tests/next_tests/integration_tests/feature_tests/iterator_tests/test_program.py	Keep iterator-level string offset usage (shift("Ioff", ...)) but remove FieldOffset declaration.
tests/next_tests/integration_tests/feature_tests/ffront_tests/test_where.py	Replace Koff[-1] usage with KDim - 1.
tests/next_tests/integration_tests/feature_tests/ffront_tests/test_program.py	Replace Ioff[1] with IDim + 1.
tests/next_tests/integration_tests/feature_tests/ffront_tests/test_gt4py_builtins.py	Migrate shifts from Ioff/Joff/Koff to Dim ± n.
tests/next_tests/integration_tests/feature_tests/ffront_tests/test_execution.py	Migrate shift tests to Dim ± n; keep explicit FieldOffsets only for dynamic as_offset paths.
tests/next_tests/integration_tests/feature_tests/ffront_tests/test_decorator.py	Remove simple_cartesian_grid dependency; adjust offset-provider cache test input.
tests/next_tests/integration_tests/feature_tests/ffront_tests/ffront_test_utils.py	Remove Ioff/Joff/Koff exports/definitions; make simple_cartesian_grid() offset_provider empty.
tests/next_tests/integration_tests/cases.py	Remove Ioff/Joff/Koff imports from shared case definitions.
tests/next_tests/fixtures/past_common.py	Remove Ioff/Joff fixture definitions as tests switch to cartesian shift syntax.
src/gt4py/next/program_processors/runners/roundtrip.py	Add CartesianOffset rendering and ensure implicit offsets are declared in generated module.
src/gt4py/next/program_processors/runners/dace/lowering/gtir_dataflow.py	Teach DaCe lowering to handle shift(CartesianOffset, ...) as cartesian shift with kind-correct dimension.
src/gt4py/next/program_processors/codegens/gtfn/itir_to_gtfn_ir.py	Define tags for dimensions referenced by CartesianOffset; render CartesianOffset as dimension tag literal.
src/gt4py/next/iterator/type_system/type_synthesizer.py	Extend dimension resolution and shift type synthesis to accept CartesianOffset.
src/gt4py/next/iterator/type_system/type_specifications.py	Add CartesianOffsetType for type inference/synthesis.
src/gt4py/next/iterator/type_system/inference.py	Add inference for itir.CartesianOffset nodes.
src/gt4py/next/iterator/transforms/trace_shifts.py	Allow tracing shift(...) sequences that include CartesianOffset.
src/gt4py/next/iterator/pretty_printer.py	Add pretty-printer support for CartesianOffset.
src/gt4py/next/iterator/ir.py	Introduce CartesianOffset IR node and hashing support.
src/gt4py/next/iterator/ir_utils/misc.py	Add helper to reconstruct common.Dimension (incl. kind) from AxisLiteral.
src/gt4py/next/iterator/ir_utils/ir_makers.py	Add im.cartesian_offset(domain, codomain) helper.
src/gt4py/next/iterator/ir_utils/domain_utils.py	Add domain translation support for cartesian shifts encoded via CartesianOffset.
src/gt4py/next/ffront/foast_to_gtir.py	Lower field(Dim ± n) to shift(CartesianOffset(...), n) using dimension type info.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 32 out of 32 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 33 out of 33 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 33 out of 33 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 34 out of 34 changed files in this pull request and generated 1 comment.

[tehrengruber]

Note for the discussion: The ITIR test changes are missing, feels like they should be in this PR.

[tehrengruber]

Is this actually a good name?

[havogt]

You mean it should be something like domain_dim, codomain_dim?

[tehrengruber]

In the IR we have CartesianOffset, but here we have Dimension. That feels inconsistent. Eventually this should be CartesianConnectivity. Maybe this is a hint that this change also belongs in the PR here. Otherwise it is not easy to asses if disallowing shift(ref_to_cart_conn, distance) is a good idea.

I did not review the rest of this file.

[Copilot]

Pull request overview

Copilot reviewed 42 out of 42 changed files in this pull request and generated no new comments.