refactor: drop generated-code formatting from next, move formatter to eve.formatting

docs/development/ADRs/next/0012-GridTools_Cpp_OTF_Steps.md

@@ -10,7 +10,7 @@ tags: [backend, cpp, gridtools, bindings, otf]
 - **Updated**: 2026-09-15
 
 > [!NOTE]
-> The entry points named below have since been renamed or moved: `program_processors.formatters.gtfn.format_sourcecode` is now `format_cpp`; `program_processors.codegens.gtfn_modules.translate_program` is now the `GTFNTranslationStep` class in `program_processors.codegens.gtfn.gtfn_module`; `otf.binding.pybind.bind_source` is now `otf.binding.nanobind.create_bindings` (nanobind replaced pybind11); and `program_processors.runners.gtfn_cpu.run_gtfn` is now `program_processors.runners.gtfn.run_gtfn`. `otf.step_types` was merged into `otf.stages`, `otf.workflow.Step` is now the `otf.workflow.Workflow` protocol, and `processor_interface` was split up — `ProgramFormatter` lives in `program_processors.program_formatter`. The decision — building the GTFN backend out of composable OTF steps — is unchanged.
+> The entry points named below have since been renamed or moved: `program_processors.formatters.gtfn.format_sourcecode` is now `format_cpp`; `program_processors.codegens.gtfn_modules.translate_program` is now the `GTFNTranslationStep` class in `program_processors.codegens.gtfn.gtfn_module`; `otf.binding.pybind.bind_source` is now `otf.binding.nanobind.create_bindings` (nanobind replaced pybind11); and `program_processors.runners.gtfn_cpu.run_gtfn` is now `program_processors.runners.gtfn.run_gtfn`. `otf.step_types` was merged into `otf.stages`, `otf.workflow.Step` is now the `otf.workflow.Workflow` protocol, and `processor_interface` was split up — `ProgramFormatter` lives in `program_processors.program_formatter`. The decision — building the GTFN backend out of composable OTF steps — is unchanged. `SourceCodeSpec` no longer carries formatting options, and generated code is no longer auto-formatted in `gt4py.next`.
 
 ## Context
 

pyproject.toml

@@ -43,6 +43,8 @@ profiling = [
 ]
 scripts = ["pyyaml>=6.0.1", "typer>=0.16.0", "packaging"]
 test = [
+  'black>=25.11',
+  'clang-format>=18.1',
   'hypothesis>=6.0.0',
   'nbmake>=1.4.6',
   'nox>=2025.02.09',
@@ -90,7 +92,6 @@ classifiers = [
 dependencies = [
   'attrs>=21.3',
   'array-api-compat>=1.13',
-  'black>=25.11',
   'boltons>=20.1',
   'cached-property>=1.5.1',
   'click>=8.0.0',
@@ -142,7 +143,7 @@ readme = 'README.md'
 requires-python = '>=3.12, <3.15, !=3.13.10, !=3.14.1'
 
 [project.optional-dependencies]
-cartesian = ['gt4py[jax,standard,testing]']
+cartesian = ['black>=25.11', 'clang-format>=18.1', 'gt4py[jax,standard,testing]']
 cuda12 = ['cupy-cuda12x>=12.0']
 cuda13 = ['cupy-cuda13x>=14.0']
 jax = [
@@ -154,7 +155,7 @@ jax-cuda13 = ['jax[cuda13-local]>=0.7.0', 'gt4py[cuda13]']
 next = ['gt4py[jax,standard,testing]']
 rocm6 = ['cupy>=13.4.1,<14.0']
 rocm7 = ['cupy-rocm-7-0>=14.0']
-standard = ['clang-format>=18.1', 'scipy>=1.16.1']
+standard = ['scipy>=1.16.1']
 testing = ['hypothesis>=6.93', 'pytest>=7.0']
 
 [project.urls]

src/gt4py/cartesian/backend/dace_backend.py

@@ -44,7 +44,6 @@
 from gt4py.cartesian.gtc.passes.oir_optimizations.utils import compute_fields_extents
 from gt4py.cartesian.gtc.passes.oir_pipeline import DefaultPipeline
 from gt4py.cartesian.utils import shash
-from gt4py.eve import codegen
 from gt4py.eve.codegen import MakoTemplate as as_mako
 from gt4py.storage.cartesian import layout, layout_registry
 
@@ -510,7 +509,7 @@ def __call__(self) -> dict[str, dict[str, str]]:
 
         implementation = DaCeComputationCodegen.apply(self.backend.builder, sdfg)
 
-        bindings = DaCeBindingsCodegen.apply(sdfg, self.module_name, backend=self.backend)
+        bindings = DaCeBindingsCodegen(self.backend).generate_sdfg_bindings(sdfg, self.module_name)
 
         bindings_ext = "cu" if self.backend.storage_info["device"] == "gpu" else "cpp"
         return {
@@ -664,7 +663,7 @@ def apply(cls, builder: StencilBuilder, sdfg: SDFG) -> str:
             state_suffix=config.Config.get("compiler.codegen_state_struct_suffix"),
         )
         computations = cls._postprocess_dace_code(code_objects, is_gpu)
-        generated_code = f"""\
+        return f"""\
 #include <gridtools/sid/sid_shift_origin.hpp>
 #include <gridtools/sid/allocator.hpp>
 #include <gridtools/stencil/cartesian.hpp>
@@ -676,11 +675,6 @@ def apply(cls, builder: StencilBuilder, sdfg: SDFG) -> str:
 {interface}
 """
 
-        if builder.options.format_source:
-            generated_code = codegen.format_source("cpp", generated_code, style="LLVM")
-
-        return generated_code
-
     def generate_dace_args(self, stencil_ir: gtir.Stencil, sdfg: SDFG) -> list[str]:
         oir = GTIRToOIR().visit(stencil_ir)
         field_extents = compute_fields_extents(oir, add_k=True)
@@ -857,13 +851,6 @@ def generate_sdfg_bindings(self, sdfg: SDFG, module_name: str) -> str:
             sid_params=self.generate_sid_params(sdfg),
         )
 
-    @classmethod
-    def apply(cls, sdfg: SDFG, module_name: str, *, backend: BaseDaceBackend) -> str:
-        generated_code = cls(backend).generate_sdfg_bindings(sdfg, module_name)
-        if backend.builder.options.format_source:
-            generated_code = codegen.format_source("cpp", generated_code, style="LLVM")
-        return generated_code
-
 
 class DaCePyExtModuleGenerator(PyExtModuleGenerator):
     def __init__(self, builder: StencilBuilder) -> None:

src/gt4py/cartesian/backend/debug_backend.py

@@ -16,7 +16,7 @@
 from gt4py.cartesian.gtc.debug.debug_codegen import DebugCodeGen
 from gt4py.cartesian.gtc.gtir_to_oir import GTIRToOIR
 from gt4py.cartesian.gtc.passes import oir_optimizations
-from gt4py.eve import codegen
+from gt4py.eve import formatting
 from gt4py.storage import layout
 from gt4py.storage.cartesian import layout_registry
 
@@ -45,7 +45,7 @@ def _generate_computation(self) -> dict[str, str | dict]:
         source_code = DebugCodeGen().visit(oir)
 
         if self.builder.options.format_source:
-            source_code = codegen.format_source("python", source_code)
+            source_code = formatting.format_python_source(source_code)
 
         caching = self.builder.caching
         computation_name = f"{caching.module_prefix}computation{caching.module_postfix}.py"

src/gt4py/cartesian/backend/gtc_common.py

@@ -19,7 +19,7 @@
 from gt4py.cartesian.backend.module_generator import BaseModuleGenerator, ModuleData
 from gt4py.cartesian.gtc import gtir, utils as gtc_utils
 from gt4py.cartesian.gtc.passes.oir_pipeline import OirPipeline
-from gt4py.eve import codegen
+from gt4py.eve import codegen, formatting
 
 
 if TYPE_CHECKING:
@@ -273,6 +273,10 @@ def _make_extension_sources(self) -> dict[str, dict[str, str]]:
         )
         gt_pyext_generator = self.PYEXT_GENERATOR_CLASS(class_name, module_name, self)
         gt_pyext_sources = gt_pyext_generator()
+        if self.builder.options.format_source:
+            for sources in gt_pyext_sources.values():
+                for file_name, source in sources.items():
+                    sources[file_name] = formatting.format_cpp_source(source)
         final_ext = ".cu" if self.languages and self.languages["computation"] == "cuda" else ".cpp"
         comp_src = gt_pyext_sources["computation"]
         for key in [k for k in comp_src.keys() if k.endswith(".src")]:

src/gt4py/cartesian/backend/gtcpp_backend.py

@@ -45,15 +45,11 @@ def __call__(self) -> dict[str, dict[str, str]]:
         )
         oir_node = oir_pipeline.run(base_oir)
         gtcpp_ir = OIRToGTCpp().visit(oir_node)
-        format_source = self.backend.builder.options.format_source
         implementation = gtcpp_codegen.GTCppCodegen.apply(
-            gtcpp_ir, gt_backend_t=self.backend.GT_BACKEND_T, format_source=format_source
+            gtcpp_ir, gt_backend_t=self.backend.GT_BACKEND_T
         )
         bindings = GTCppBindingsCodegen.apply(
-            gtcpp_ir,
-            module_name=self.module_name,
-            backend=self.backend,
-            format_source=format_source,
+            gtcpp_ir, module_name=self.module_name, backend=self.backend
         )
         bindings_ext = ".cu" if self.backend.GT_BACKEND_T == "gpu" else ".cpp"
         return {
@@ -115,11 +111,7 @@ def visit_Program(self, node: gtcpp.Program, **kwargs):
 
     @classmethod
     def apply(cls, root, *, module_name="stencil", **kwargs) -> str:
-        generated_code = cls(kwargs.get("backend")).visit(root, module_name=module_name, **kwargs)
-        if kwargs.get("format_source", True):
-            generated_code = codegen.format_source("cpp", generated_code, style="LLVM")
-
-        return generated_code
+        return cls(kwargs.get("backend")).visit(root, module_name=module_name, **kwargs)
 
 
 class GTBaseBackend(BaseGTBackend):

src/gt4py/cartesian/backend/module_generator.py

@@ -25,7 +25,7 @@
 from gt4py.cartesian.gtc.passes.oir_access_kinds import compute_access_kinds
 from gt4py.cartesian.gtc.passes.oir_optimizations.utils import compute_fields_extents
 from gt4py.cartesian.gtc.utils import dimension_flags_to_names
-from gt4py.eve import codegen
+from gt4py.eve import formatting
 
 
 if TYPE_CHECKING:
@@ -107,7 +107,6 @@ def make_args_data_from_gtir(pipeline: GtirPipeline) -> ModuleData:
 
 
 class BaseModuleGenerator(abc.ABC):
-    SOURCE_LINE_LENGTH = 120
     TEMPLATE_INDENT_SIZE = 4
     TEMPLATE_RESOURCE = "stencil_module.py.in"
 
@@ -149,10 +148,8 @@ def __call__(self, args_data: ModuleData) -> str:
             post_run=self.generate_post_run(),
             implementation=self.generate_implementation(),
         )
-        if self.builder.options.as_dict()["format_source"]:
-            module_source = codegen.format_source(
-                "python", module_source, line_length=self.SOURCE_LINE_LENGTH
-            )
+        if self.builder.options.format_source:
+            module_source = formatting.format_python_source(module_source)
 
         return module_source
 
@@ -202,16 +199,11 @@ def generate_backend_name(self) -> str:
 
     def generate_sources(self) -> dict[str, str]:
         """
-        Return the source code of the stencil definition in string format.
+        Return the source code of the stencil definition verbatim, in string format.
 
         This is unlikely to require overriding.
         """
-        if self.builder.gtir.sources is not None:
-            return {
-                key: codegen.format_source("python", value, line_length=self.SOURCE_LINE_LENGTH)
-                for key, value in self.builder.gtir.sources.items()
-            }
-        return {}
+        return dict(self.builder.gtir.sources or {})
 
     def generate_constants(self) -> dict[str, str]:
         """

src/gt4py/cartesian/backend/numpy_backend.py

@@ -16,7 +16,7 @@
 from gt4py.cartesian.gtc.gtir_to_oir import GTIRToOIR
 from gt4py.cartesian.gtc.numpy import npir
 from gt4py.cartesian.gtc.passes import oir_optimizations as oir_opt
-from gt4py.eve import codegen
+from gt4py.eve import formatting
 from gt4py.storage import layout
 from gt4py.storage.cartesian import layout_registry
 
@@ -44,7 +44,7 @@ def generate_computation(self) -> dict[str, str | dict]:
         source = numpy.NpirCodegen.apply(self.npir, ignore_np_errstate=ignore_np_errstate)
 
         if self.builder.options.format_source:
-            source = codegen.format_source("python", source)
+            source = formatting.format_python_source(source)
 
         caching = self.builder.caching
         computation_name = f"{caching.module_prefix}computation{caching.module_postfix}.py"

src/gt4py/cartesian/gtc/gtcpp/gtcpp_codegen.py

@@ -324,8 +324,4 @@ def apply(cls, root: LeafNode, **kwargs: Any) -> str:
             raise ValueError("apply() requires gtcpp.Progam root node")
         if "gt_backend_t" not in kwargs:
             raise TypeError("apply() missing 1 required keyword-only argument: 'gt_backend_t'")
-        generated_code = super().apply(root, offset_limit=_offset_limit(root), **kwargs)
-        if kwargs.get("format_source", True):
-            generated_code = codegen.format_source("cpp", generated_code, style="LLVM")
-
-        return generated_code
+        return super().apply(root, offset_limit=_offset_limit(root), **kwargs)

src/gt4py/eve/__init__.py

@@ -11,7 +11,7 @@
 The internal dependencies between modules are the following (each module depends
 on some of the previous ones):
 
-  0. xtyping
+  0. formatting, xtyping
   1. exceptions, pattern_matching, type_definitions
   2. utils
   3. type_validation