bench: add libcst visitor benchmarks for multi-file and full pipeline

- test_benchmark_libcst_multi_file: discover_functions + get_code_optimization_context across 10 real source files
- test_benchmark_libcst_pipeline: full discover → extract → replace → merge pipeline on one file

tests/benchmarks/test_benchmark_libcst_multi_file.py

@@ -0,0 +1,75 @@
+"""Benchmark libcst visitor performance across many files.
+
+Exercises the visitor-heavy codepaths that benefit from the libcst dispatch
+table cache: discover_functions + get_code_optimization_context on multiple
+real source files.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+from codeflash.discovery.functions_to_optimize import FunctionToOptimize
+from codeflash.languages.python.context.code_context_extractor import get_code_optimization_context
+from codeflash.languages.python.support import PythonSupport
+from codeflash.models.models import FunctionParent
+
+# Real source files from the codeflash codebase, chosen for size and visitor diversity.
+_CODEFLASH_ROOT = Path(__file__).parent.parent.parent.resolve() / "codeflash"
+
+_SOURCE_FILES: list[Path] = [
+    _CODEFLASH_ROOT / "languages" / "function_optimizer.py",
+    _CODEFLASH_ROOT / "languages" / "python" / "context" / "code_context_extractor.py",
+    _CODEFLASH_ROOT / "languages" / "python" / "support.py",
+    _CODEFLASH_ROOT / "languages" / "python" / "static_analysis" / "code_extractor.py",
+    _CODEFLASH_ROOT / "languages" / "python" / "static_analysis" / "code_replacer.py",
+    _CODEFLASH_ROOT / "code_utils" / "instrument_existing_tests.py",
+    _CODEFLASH_ROOT / "benchmarking" / "compare.py",
+    _CODEFLASH_ROOT / "models" / "models.py",
+    _CODEFLASH_ROOT / "discovery" / "discover_unit_tests.py",
+    _CODEFLASH_ROOT / "languages" / "base.py",
+]
+
+# For each file, pick one top-level function to extract context for.
+# (class, function_name) — class=None means module-level.
+_TARGETS: list[tuple[Path, str | None, str]] = [
+    (_SOURCE_FILES[0], "FunctionOptimizer", "replace_function_and_helpers_with_optimized_code"),
+    (_SOURCE_FILES[1], None, "get_code_optimization_context"),
+    (_SOURCE_FILES[2], "PythonSupport", "discover_functions"),
+    (_SOURCE_FILES[3], None, "add_global_assignments"),
+    (_SOURCE_FILES[4], None, "replace_functions_in_file"),
+    (_SOURCE_FILES[5], None, "inject_profiling_into_existing_test"),
+    (_SOURCE_FILES[6], None, "compare_branches"),
+    (_SOURCE_FILES[7], None, "get_comment_prefix"),
+    (_SOURCE_FILES[8], None, "discover_unit_tests"),
+    (_SOURCE_FILES[9], None, "convert_parents_to_tuple"),
+]
+
+
+def _discover_all() -> None:
+    """Run discover_functions on all source files."""
+    ps = PythonSupport()
+    for file_path in _SOURCE_FILES:
+        source = file_path.read_text()
+        ps.discover_functions(source=source, file_path=file_path)
+
+
+def _extract_all_contexts() -> None:
+    """Run get_code_optimization_context on every target function."""
+    project_root = _CODEFLASH_ROOT.parent
+    for file_path, class_name, func_name in _TARGETS:
+        parents = [FunctionParent(name=class_name, type="ClassDef")] if class_name else []
+        fto = FunctionToOptimize(
+            function_name=func_name, file_path=file_path, parents=parents, starting_line=None, ending_line=None
+        )
+        get_code_optimization_context(fto, project_root)
+
+
+def test_benchmark_discover_functions_multi_file(benchmark) -> None:
+    """Discover functions across 10 source files."""
+    benchmark(_discover_all)
+
+
+def test_benchmark_extract_context_multi_file(benchmark) -> None:
+    """Extract code optimization context for 10 functions across 10 files."""
+    benchmark(_extract_all_contexts)

tests/benchmarks/test_benchmark_libcst_pipeline.py

@@ -0,0 +1,56 @@
+"""Benchmark the full libcst-heavy pipeline on a single file.
+
+Runs discover → extract context → replace functions → add global assignments
+in sequence, exercising ~15 distinct visitor/transformer classes in one pass.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+from codeflash.discovery.functions_to_optimize import FunctionToOptimize
+from codeflash.languages.python.context.code_context_extractor import get_code_optimization_context
+from codeflash.languages.python.static_analysis.code_extractor import add_global_assignments
+from codeflash.languages.python.static_analysis.code_replacer import replace_functions_in_file
+from codeflash.languages.python.support import PythonSupport
+
+_CODEFLASH_ROOT = Path(__file__).parent.parent.parent.resolve() / "codeflash"
+_PROJECT_ROOT = _CODEFLASH_ROOT.parent
+
+# Target: a real, non-trivial file with classes and module-level functions.
+_TARGET_FILE = _CODEFLASH_ROOT / "languages" / "python" / "static_analysis" / "code_extractor.py"
+_TARGET_FUNC = "add_global_assignments"
+
+# A second file to serve as "optimized" source for replace/merge steps.
+_SECOND_FILE = _CODEFLASH_ROOT / "languages" / "python" / "static_analysis" / "code_replacer.py"
+
+
+def _run_pipeline() -> None:
+    """Simulate a single-file optimization pass through the full visitor pipeline."""
+    source = _TARGET_FILE.read_text()
+    source2 = _SECOND_FILE.read_text()
+
+    # 1. Discover functions (FunctionVisitor + MetadataWrapper)
+    ps = PythonSupport()
+    functions = ps.discover_functions(source=source, file_path=_TARGET_FILE)
+
+    # 2. Extract code optimization context (multiple collectors + dependency resolver)
+    fto = FunctionToOptimize(
+        function_name=_TARGET_FUNC, file_path=_TARGET_FILE, parents=[], starting_line=None, ending_line=None
+    )
+    get_code_optimization_context(fto, _PROJECT_ROOT)
+
+    # 3. Replace functions (GlobalFunctionCollector + GlobalFunctionTransformer)
+    # Use a class method from discovered functions if available, else module-level.
+    func_names = [_TARGET_FUNC]
+    replace_functions_in_file(
+        source_code=source, original_function_names=func_names, optimized_code=source2, preexisting_objects=set()
+    )
+
+    # 4. Add global assignments (6 visitors/transformers)
+    add_global_assignments(source2, source)
+
+
+def test_benchmark_full_pipeline(benchmark) -> None:
+    """Full discover → extract → replace → merge pipeline on one file."""
+    benchmark(_run_pipeline)