fix the tracer

[KRRT7]

PR Type

Bug fix, Enhancement

---

Description

• Parse args preserving unknown for target

• Set PYTHONPATH to project root

• Fix pytest_split return shape

• Skip license bump on dev/post builds

---

Diagram Walkthrough

flowchart LR
  ARGS["Args parsing"] -- "parse_known_args; keep unknown" --> TRACER["Tracer main()"]
  TRACER -- "spawn subprocess" --> SUBP["tracing_new_process.py"]
  ENV["Project root in PYTHONPATH"] -- "env=..." --> SUBP
  SPLIT["pytest_split()"] -- "ensure list of lists" --> PARALLEL["Parallel test splits"]
  VERSION["update_license_version"] -- "skip dev/post versions" --> LICENSE["License version unchanged"]

Loading

File Walkthrough

	Relevant files
Enhancement	tracer.py Robust arg handling and PYTHONPATH for tracer subprocesses codeflash/tracer.py Use parse_known_args; preserve unknowns in sys.argv. Respect --module via parsed temp args. Inject project root into PYTHONPATH for subprocesses. Pass env to Popen/run for correct imports. +23/-2    update_license_version.py Skip license version update for non-final builds                  codeflash/update_license_version.py Skip updates for dev, local (+), and post releases. Guard to avoid unintended license version bumps. +3/-0
Bug fix	pytest_parallelization.py Fix pytest_split return structure for small sets                  codeflash/tracing/pytest_parallelization.py Ensure return is list of lists when <4 files. Prevent shape mismatch in downstream consumers. +1/-1

---

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Arg Parsing Behavior Rewriting sys.argv with only unknown args after parse_known_args may drop flags the subprocess expects (e.g., the parsed --module or other known options). Validate that downstream logic and help/usage still behave correctly and that required args are not lost. temp_parsed, unknown_args = parser.parse_known_args() parsed_args.module = temp_parsed.module sys.argv[:] = unknown_args Env Composition PYTHONPATH prefixing is duplicated and string-based; confirm no duplicates or path length issues and that project_root is correct for both parallel and single-process paths. Consider normalizing to avoid repeated prefixes on nested invocations. env = os.environ.copy() pythonpath = env.get("PYTHONPATH", "") project_root_str = str(project_root) if pythonpath: env["PYTHONPATH"] = f"{project_root_str}{os.pathsep}{pythonpath}" else: env["PYTHONPATH"] = project_root_str Return Shape Consistency Changing the zero-split case to return [test_files] aligns shapes, but verify all callers handle an empty test_files list and that num_splits reduction logic still yields expected distribution when len<4. return [test_files], test_paths

[github-actions]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
Possible issue	Preserve argv[0] when rewriting argv Preserve the original sys.argv[0] (script path) when replacing argv, otherwise downstream command construction may break. Insert the original first argument back to sys.argv when reassigning. codeflash/tracer.py [69-71] temp_parsed, unknown_args = parser.parse_known_args() parsed_args.module = temp_parsed.module -sys.argv[:] = unknown_args +# Preserve original argv[0] (executable/script) while passing unknown args through +original_argv0 = sys.argv[0] if sys.argv else "" +sys.argv[:] = [original_argv0, *unknown_args] if original_argv0 else unknown_args Suggestion importance[1-10]: 7 __ Why: Correctly identifies that replacing sys.argv with only unknown args drops argv[0], which can break downstream command construction; the proposed fix is small, accurate, and low-risk.	Medium
	Handle empty test list edge-case Ensure the function consistently returns a tuple where the first element is a list of lists. When test_files is empty, return an empty list of splits instead of [[]] to avoid running an empty split. codeflash/tracing/pytest_parallelization.py [63-65] max_possible_splits = len(test_files) // 4 if max_possible_splits == 0: - return [test_files], test_paths + return ([], test_paths) if not test_files else ([test_files], test_paths) Suggestion importance[1-10]: 5 __ Why: This is a reasonable edge-case improvement to return no splits when there are no tests, aligning return type consistency; impact is modest and context-dependent.	Low
General	Prevent PYTHONPATH duplication Avoid duplicating project_root in PYTHONPATH when it’s already present, which can lead to very long env vars over multiple runs. Check containment before prepending. codeflash/tracer.py [133-139] env = os.environ.copy() pythonpath = env.get("PYTHONPATH", "") project_root_str = str(project_root) -if pythonpath: - env["PYTHONPATH"] = f"{project_root_str}{os.pathsep}{pythonpath}" +path_elems = pythonpath.split(os.pathsep) if pythonpath else [] +if project_root_str not in path_elems: + env["PYTHONPATH"] = os.pathsep.join([project_root_str] + path_elems) if path_elems else project_root_str else: - env["PYTHONPATH"] = project_root_str + env["PYTHONPATH"] = pythonpath Suggestion importance[1-10]: 6 __ Why: Avoiding repeated project_root entries in PYTHONPATH is a practical enhancement that prevents env var bloat across runs; the check is straightforward and accurate.	Low

[KRRT7]

previously pytest_split was just a string, when a list of list was expected, it ended up splitting that string into a list of individual strings causing the parser to fail