warntrace 0.1.0.post1

Find which package caused a Python warning and where your code triggered it.

Warntrace

Trace the origin of Python warnings — classify them by ownership (application, direct dependency, transitive, stdlib), suppress noise, and fail CI when policy is violated.

Installation

uv add --dev warntrace
# or
pip install warntrace

Requires Python >= 3.10.

Quick start

# Capture warnings from any Python command
warntrace run python your_script.py

# Capture warnings from your test suite
warntrace run pytest

Warntrace intercepts every warnings.warn() call in the child process, classifies each warning by origin, and writes a JSON report to stdout after the child exits.

CLI reference

warntrace run [options] -- <command> [args...]

Flag	Default	Description
--format	json	Output format: json, detailed, or summary
--output FILE	stdout	Write report to a file instead of stdout
--max-frames N	unlimited	Limit stack frames per warning in terminal output
--show-all	off	Capture normally-suppressed warnings (e.g. BytesWarning)
--no-passthrough	off	Suppress original warning output from the child process
--fail-on POLICY	off	Exit code 3 when a warning matches the policy (see below)
--root PATH	child's CWD	Application root directory for classification
--no-color	off	Disable ANSI color in terminal output

Examples

# Basic — run a script and see the default JSON report
warntrace run python -c "import warnings; warnings.warn('hello')"

# Detailed terminal output with limited stack frames
warntrace run --format detailed --max-frames 5 pytest

# Summary — one line per warning with total counts
warntrace run --format summary pytest

# Write report to a file (no ANSI codes in file output)
warntrace run --output report.json pytest

# Collect all warnings silently, even normally-suppressed ones
warntrace run --no-passthrough --show-all pytest

# Fail CI when application-origin warnings are found
warntrace run --fail-on application pytest

# Treat triggered dependency warnings as application-level too
warntrace run --fail-on application pytest

# Fail on any warning (application + deps + stdlib + unknown)
warntrace run --fail-on all pytest

# Opt out of policy failure explicitly
warntrace run --fail-on none pytest

# Specify granular origins to fail on
warntrace run --fail-on direct_dependency transitive_dependency pytest

# Combined: silent, fail on app warnings, save report
warntrace run --no-passthrough --fail-on application --output ci-report.json pytest

Python API

from warntrace import capture_warnings

with capture_warnings() as tracer:
    import warnings
    warnings.warn("something is deprecated", DeprecationWarning)

report = tracer.stop()
print(report.to_dict())

The context manager installs the capture hook, runs your code, restores the original warning state, and returns a classified WarningReport. The report contains the full warning list with origin, stack, dependency paths, and occurrence counts.

Lower-level API

from warntrace import WarningTracer

tracer = WarningTracer(show_all=True, passthrough=False)
tracer.start()
# ... your code ...
report = tracer.stop()
print(report.to_dict())

Configuration

Parameter	Default	Description
root	Path.cwd()	Application root for origin classification
show_all	False	Capture normally-suppressed warning types
passthrough	True	Forward warnings to the original handler

Output formats

JSON (default)

{
  "schema_version": "0.1",
  "summary": {
    "unique_warnings": 2,
    "total_occurrences": 5,
    "application": 1,
    "direct_dependency": 1,
    "transitive_dependency": 0,
    "standard_library": 0,
    "unknown": 0
  },
  "warnings": [
    {
      "category": "DeprecationWarning",
      "message": "use foobar() instead of deprecated_func()",
      "origin": "direct_dependency",
      "triggered_directly_by_application": true,
      "occurrences": 3,
      "emitted_from": {
        "filename": "/path/to/site-packages/oldpkg/utils.py",
        "lineno": 42,
        "distribution": {
          "name": "oldpkg",
          "normalized_name": "oldpkg",
          "version": "1.2.3"
        }
      },
      "application_frame": {
        "filename": "/home/user/project/app.py",
        "lineno": 15,
        "function": "run",
        "source_line": "oldpkg.deprecated_func()"
      },
      "dependency_path": [
        {"name": "myproject", "normalized_name": "myproject", "version": null},
        {"name": "oldpkg", "normalized_name": "oldpkg", "version": "1.2.3"}
      ],
      "stack": [
        {"filename": "/home/user/project/app.py", "lineno": 15, "function": "run", "source_line": "oldpkg.deprecated_func()"},
        {"filename": "/path/to/site-packages/oldpkg/utils.py", "lineno": 42, "function": "deprecated_func", "source_line": "warnings.warn(...)"}
      ]
    }
  ]
}

Detailed terminal output

warntrace report — 1 unique warning, 3 total occurrences

─── Warning 1 of 1 ───────────────────────────────────────────────────────────
  Origin:        !  direct_dependency (triggered by your code)
  Category:      DeprecationWarning
  Message:       use foobar() instead of deprecated_func()
  Location:      oldpkg/utils.py:42
  Occurrences:   3
  Application:
    myproject/app.py:15  run()
      -> oldpkg.deprecated_func()
  Dependency path:
    myproject → oldpkg 1.2.3
  Stack:
     1  myproject/app.py:15  run()
     2  oldpkg/utils.py:42  deprecated_func()

Summary output

  application       1  DeprecationWarning: use foobar() instead...
  direct_dependency  3  DeprecationWarning: use foobar() instead...
────────────────────────────────────────────────────────
warntrace report — 2 unique warnings, 4 total occurrences

Classification

Warntrace assigns an origin to every captured warning by walking the stack and mapping each frame to a package or to the Python standard library. There are five origin levels:

Origin	Meaning
application	The warning was emitted from your own code (under the project root)
direct_dependency	The warning was emitted by a package listed in your dependencies
transitive_dependency	The warning was emitted by a dependency of one of your dependencies
standard_library	The warning was emitted from CPython stdlib
unknown	The file could not be mapped to any known package

Additionally, each warning has a triggered_directly_by_application boolean. This is true when an application frame appears in the call stack below the warning emission — meaning your code called the code that warned, even if the warning itself comes from a dependency.

See docs/CLASSIFICATION.md for the detailed pipeline documentation.

CI integration

Policies are expressed via --fail-on:

Policy	Warnings that trigger failure
none	Never (explicit opt-out)
application	Origin is application or triggered_directly_by_application is true
direct	Origin is application or direct_dependency
all	Any captured warning
(individual origins)	direct_dependency, transitive_dependency, standard_library, unknown

Exit codes:

Code	Meaning
0	Success — no policy failure
2	Warntrace usage error (e.g. missing command)
3	Policy matched — warnings exceed the configured threshold

Child-process failure (e.g. pytest test failures) takes priority over policy: if the child exits non-zero, that exit code is returned even when a policy would have matched.

Example: CI workflow

# .github/workflows/ci.yml
- run: warntrace run --no-passthrough --fail-on application --output warnings.json pytest

The test suite runs normally. If any application-origin or application-triggered warnings are found, the step exits with code 3. The full report is saved to warnings.json for inspection.

Privacy

Warntrace makes no network requests. All analysis is local. It reads:

• Your project's dependency metadata (importlib.metadata)
• The filesystem path of each stack frame
• Environment variables (for configuration only)

No data is collected, logged, or transmitted.

Known limitations

• Pytest stack frames. Under pytest, captured stacks include ~30 frames of pytest/pluggy internals. Use --max-frames to limit terminal output.
• Self-testing. Running warntrace run pytest on warntrace's own test suite reports 0 warnings because the test files manipulate capture state at import time. This does not affect normal projects.
• Dependency graph accuracy. The graph is built from installed package metadata (Requires-Dist). Optional/extra dependencies are excluded. Namespace packages with ambiguous ownership return unknown.