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',
   'factory-boy>=3.3.3',
   'hypothesis>=6.0.0',
   'nbmake>=1.4.6',
@@ -91,7 +93,6 @@ 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

@@ -45,7 +45,7 @@
 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 import formatting
 from gt4py.eve.codegen import MakoTemplate as as_mako
 from gt4py.storage.cartesian import layout, layout_registry
 
@@ -679,7 +679,7 @@ def apply(cls, builder: StencilBuilder, sdfg: SDFG) -> str:
 """
 
         if builder.options.format_source:
-            generated_code = codegen.format_source("cpp", generated_code, style="LLVM")
+            generated_code = formatting.format_cpp_source(generated_code)
 
         return generated_code
 
@@ -863,7 +863,7 @@ def generate_sdfg_bindings(self, sdfg: SDFG, module_name: str) -> str:
     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")
+            generated_code = formatting.format_cpp_source(generated_code)
         return generated_code
 
 

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/gtcpp_backend.py

@@ -24,7 +24,7 @@
 from gt4py.cartesian.gtc.gtcpp.oir_to_gtcpp import OIRToGTCpp
 from gt4py.cartesian.gtc.gtir_to_oir import GTIRToOIR
 from gt4py.cartesian.gtc.passes.oir_pipeline import DefaultPipeline
-from gt4py.eve import codegen
+from gt4py.eve import codegen, formatting
 from gt4py.storage.cartesian import layout, layout_registry
 
 
@@ -117,7 +117,7 @@ def visit_Program(self, node: gtcpp.Program, **kwargs):
     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")
+            generated_code = formatting.format_cpp_source(generated_code)
 
         return generated_code
 

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:
@@ -150,8 +150,8 @@ def __call__(self, args_data: ModuleData) -> str:
             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
+            module_source = formatting.format_python_source(
+                module_source, line_length=self.SOURCE_LINE_LENGTH
             )
 
         return module_source
@@ -208,7 +208,7 @@ def generate_sources(self) -> dict[str, str]:
         """
         if self.builder.gtir.sources is not None:
             return {
-                key: codegen.format_source("python", value, line_length=self.SOURCE_LINE_LENGTH)
+                key: formatting.format_python_source(value, line_length=self.SOURCE_LINE_LENGTH)
                 for key, value in self.builder.gtir.sources.items()
             }
         return {}

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

@@ -20,7 +20,7 @@
     UnaryOperator,
 )
 from gt4py.cartesian.gtc.gtcpp import gtcpp
-from gt4py.eve import codegen
+from gt4py.eve import codegen, formatting
 from gt4py.eve.codegen import FormatTemplate as as_fmt, MakoTemplate as as_mako
 from gt4py.eve.concepts import LeafNode
 
@@ -326,6 +326,6 @@ def apply(cls, root: LeafNode, **kwargs: Any) -> str:
             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")
+            generated_code = formatting.format_cpp_source(generated_code)
 
         return generated_code

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

src/gt4py/eve/codegen.py

@@ -14,17 +14,14 @@
 import collections.abc
 import contextlib
 import inspect
-import os
 import re
 import string
-import subprocess
 import sys
 import textwrap
 import types
-from collections.abc import Callable, Collection, Iterator, Mapping, Sequence
+from collections.abc import Collection, Iterator, Mapping, Sequence
 from typing import Any, ClassVar, Optional, TypeVar, Union, overload
 
-import black
 import jinja2
 from mako import template as mako_tpl
 from typing_extensions import Protocol, runtime_checkable
@@ -34,24 +31,6 @@
 from .visitors import NodeVisitor
 
 
-SourceFormatter = Callable[[str], str]
-
-SOURCE_FORMATTERS: dict[str, SourceFormatter] = {}
-"""Global dict storing registered formatters."""
-
-
-class FormatterNameError(exceptions.EveRuntimeError):
-    """Run-time error registering a new source code formatter."""
-
-    ...
-
-
-class FormattingError(exceptions.EveRuntimeError):
-    """Run-time error applying a source code formatter."""
-
-    ...
-
-
 class TemplateDefinitionError(exceptions.EveTypeError):
     """Template definition error."""
 
@@ -64,111 +43,6 @@ class TemplateRenderingError(exceptions.EveRuntimeError):
     ...
 
 
-def register_formatter(language: str) -> Callable[[SourceFormatter], SourceFormatter]:
-    """Register source code formatters for specific languages (decorator)."""
-
-    def _decorator(formatter: SourceFormatter) -> SourceFormatter:
-        if language in SOURCE_FORMATTERS:
-            raise FormatterNameError(f"Another formatter for language '{language}' already exists")
-
-        assert callable(formatter)
-        SOURCE_FORMATTERS[language] = formatter
-
-        return formatter
-
-    return _decorator
-
-
-@register_formatter("python")
-def format_python_source(
-    source: str,
-    *,
-    line_length: int = 100,
-    python_versions: Optional[set[str]] = None,
-    string_normalization: bool = True,
-) -> str:
-    """Format Python source code using black formatter."""
-    python_versions = python_versions or {f"{sys.version_info.major}{sys.version_info.minor}"}
-    target_versions = set(black.TargetVersion[f"PY{v.replace('.', '')}"] for v in python_versions)  # type: ignore[attr-defined]  # .TargetVersion implicitly exported
-
-    formatted_source = black.format_str(
-        source,
-        mode=black.FileMode(
-            line_length=line_length,
-            target_versions=target_versions,
-            string_normalization=string_normalization,
-        ),
-    )
-    assert isinstance(formatted_source, str)
-
-    return formatted_source
-
-
-def _get_clang_format() -> Optional[str]:
-    """Return the clang-format executable, or None if not available."""
-    executable = os.getenv("CLANG_FORMAT_EXECUTABLE", "clang-format")
-    try:
-        assert isinstance(executable, str)
-        if subprocess.run([executable, "--version"], capture_output=True).returncode != 0:
-            return None
-    except Exception:
-        return None
-
-    return executable
-
-
-_CLANG_FORMAT_EXECUTABLE = _get_clang_format()
-
-
-if _CLANG_FORMAT_EXECUTABLE is not None:
-
-    @register_formatter("cpp")
-    def format_cpp_source(
-        source: str,
-        *,
-        style: Optional[str] = None,
-        fallback_style: Optional[str] = None,
-        sort_includes: bool = False,
-    ) -> str:
-        """Format C++ source code using clang-format."""
-        assert isinstance(_CLANG_FORMAT_EXECUTABLE, str)
-        args = [_CLANG_FORMAT_EXECUTABLE, "--assume-filename=_gt4py_generated_file.cpp"]
-        if style:
-            args.append(f"--style={style}")
-        if fallback_style:
-            args.append(f"--fallback-style={style}")
-        if sort_includes:
-            args.append("--sort-includes")
-
-        try:
-            # use a timeout as clang-format used to deadlock on some sources
-            formatted_source = subprocess.run(
-                args, check=True, input=source, capture_output=True, text=True, timeout=3
-            ).stdout
-        except subprocess.TimeoutExpired:
-            return source
-
-        assert isinstance(formatted_source, str)
-        return formatted_source
-
-
-def format_source(language: str, source: str, *, skip_errors: bool = True, **kwargs: Any) -> str:
-    """Format source code if a formatter exists for the specific language."""
-    formatter = SOURCE_FORMATTERS.get(language, None)
-    try:
-        if formatter:
-            return formatter(source, **kwargs)
-        else:
-            raise FormattingError(f"Missing formatter for '{language}' language")
-    except Exception as e:
-        if skip_errors:
-            return source
-        else:
-            raise FormattingError(
-                f"Something went wrong when trying to format '{language}' source code"
-            ) from e
-
-
 class Name:
     """Text formatter with different case styles for symbol names in source code."""