xwlazy 1.0.1.48

Convenience wrapper for exonware-xwlazy - provides 'import xwlazy' alias

xwlazy ⚡️

Install missing Python packages on first use, without changing your imports.

• Who it is for: Python projects with optional or heavy dependencies (data, ML, plugins) where you do not want to pre-install everything.
• What it does: Watches imports, installs missing packages the first time code actually uses them, and records them in your project.
• Why it is different: Per-package scope (xwsystem lazy, xwnode normal), mapping-aware install (bs4 -> beautifulsoup4), and PEP 668 and audit options built in.

---

Quick start 🚀

1. Install

pip install exonware-xwlazy
# or
pip install xwlazy

Works in both local/system Python and virtual environments (venv, virtualenv, conda, uv, etc.):

• On a system interpreter, xwlazy respects PEP 668 and will refuse to install into externally-managed environments.
• Inside a venv, it uses the active environment pip.

2. Enable for your package (one line in __init__.py)

from xwlazy import auto_enable_lazy

auto_enable_lazy(__package__)  # or auto_enable_lazy()

Then keep using normal imports. The first time a dependency is missing, xwlazy installs it; after that it behaves as a normal import.

3. Zero-code opt-in (optional)

Add this to pyproject.toml:

[project]
keywords = ["xwlazy-enabled"]

After pip install -e ., lazy mode is enabled for that project based on metadata.

---

What you get in practice ⭐

Capability	What it does
On-demand install	Missing package triggers pip install when code first touches it. No manual install step for optional features.
Per-package scope	Each package turns lazy on or off. xwsystem can be lazy, xwnode not - no cross-talk.
Keyword opt-in	"xwlazy-enabled" in pyproject.toml turns lazy on with no code change.
Two-stage load	Import time: missing imports logged, no crash. Use time: install then run. You keep normal import style.
Policy and audit	Allow/deny lists, lockfile, SBOM, and PEP 668 checks (no install into externally-managed system Python).
Persist to project	On successful install, xwlazy can add the package to requirements.txt and/or pyproject.toml. Control this with XWLAZY_PERSIST_EXTRAS and XWLAZY_NO_PERSIST.

Single implementation file: src/exonware/xwlazy.py; src/xwlazy.py re-exports. src/_old/ contains legacy/reference code only and is not shipped or imported.

---

Built-in library mappings

xwlazy ships with a curated mapping file (src/exonware/xwlazy_external_libs.toml) so common ecosystems “just work” out of the box:

• Data & ML: numpy, pandas, scipy, scikit-learn, statsmodels, xgboost, lightgbm, catboost, joblib, dask, …
• Deep learning & AI: torch, tensorflow/tf, keras, transformers, jax, jaxlib, …
• Visualization & geo: matplotlib, seaborn, plotly, bokeh, altair, graphviz, folium, geopandas, …
• Web & APIs: requests, httpx, aiohttp, fastapi, uvicorn, django, flask, starlette, …
• Formats & I/O: PyYAML, ruamel.yaml, beautifulsoup4 (bs4), pyarrow, fastavro, h5py, and more.

You can extend or override these mappings by editing that TOML file in your own project.

---

Why xwlazy? Benchmarked and mapping-aware 🏆

We run xwlazy against other lazy-import libraries (pipimport, deferred-import, lazy-loader, lazy-imports, pylazyimports, lazi, lazy-imports-lite) in a dedicated benchmark campaign. Here’s what stands out.

Performance (latest competition run)

• Medium load: 4.06 ms vs 4.54 ms next best in the benchmark campaign.
• Heavy / enterprise: 14.46 ms heavy, 41.37 ms enterprise with auto-install, per-package isolation, and audit features on.

Feature comparison

Other tools in the benchmark focus on different trade-offs: many do lazy import only (no auto-install), or auto-install but assume import name == pip name. In that comparison set, xwlazy combines: auto-install, lazy import, global import hook, mapping-aware install, pyproject/build integration, import tracing, per-package isolation, lockfile/SBOM, PEP 668 awareness, and a one-liner API. See the Library Feature Comparison table in the campaign README.

Where other lazy installers break (no mapping)

Libraries that assume import name == pip package name will fail or install the wrong thing on very common imports. Examples:

You write	Pip package to install	What happens without mapping
import bs4	beautifulsoup4	They run pip install bs4 → wrong or missing.
import yaml	PyYAML	pip install yaml → fails (no such package).
import sklearn	scikit-learn	pip install sklearn → fails.
import cv2	opencv-python	pip install cv2 → fails.
import PIL	Pillow	pip install PIL → fails.
import attr	attrs	pip install attr → wrong package.
import pandas	pandas	Same name, so may work—but import sklearn or import bs4 in the same project won’t.

xwlazy’s curated mapping (xwlazy_external_libs.toml) resolves these so that import bs4, import yaml, import sklearn, import cv2, import PIL, etc. install the correct pip package automatically. No per-import configuration or wrapper API needed.

---

Modes and strategies 🎛️

xwlazy combines when to install (lazy vs normal imports) with how to install (strategy):

Strategy	What it does	Use when
smart	Uses manifests and mappings to install missing deps on first use.	Default for most projects.
pip	Installs missing deps with plain pip under the lazy import hook.	You want vanilla pip behavior.
wheel	Prefers wheel-based installs when available.	Environments with prebuilt wheels.
cached	Reuses previously resolved install candidates across runs.	Repeat runs with similar deps.

If you don’t call auto_enable_lazy(...), imports stay normal and nothing is installed lazily.

Example:

from xwlazy import auto_enable_lazy

auto_enable_lazy("xwsystem", mode="smart")

---

Security and production 🛡️

• Deny list: a central list of blocked packages loaded from the [deny_list] section in xwlazy_external_libs.toml.
• Lockfile (opt-in): when auditing is enabled, installed packages and basic stats persist to ~/.xwlazy/xwlazy.lock.toml for reproducibility.
• Persist to project: when an install succeeds, xwlazy adds the package to your project’s requirements.txt and/or pyproject.toml (default: [project.optional-dependencies.full] if present, else [project.dependencies]). Override with XWLAZY_PERSIST_EXTRAS=<name> (write to that extras group) or XWLAZY_PERSIST_EXTRAS=none (force dependencies). Disable with XWLAZY_NO_PERSIST=1.
• SBOM (opt-in): when auditing is enabled, generate_sbom() writes ~/.xwlazy/xwlazy_sbom.toml so you can audit what was installed and when.
• Async I/O (default): file updates (persist-to-project, lockfile, audit log) run in a background worker so imports/installs don’t block your app. Set XWLAZY_ASYNC_IO=0 to force synchronous writes.
• PEP 668: xwlazy won’t install into externally-managed environments; it will tell you to use a venv instead.

Auditing is disabled by default. To enable lockfile/SBOM writes, set XWLAZY_AUDIT_ENABLED=1 in the environment before importing xwlazy.

For production, you typically pre-install pinned dependencies, keep the lazy hook in smart or pip strategy for edge cases, and (optionally) rely on the lockfile/SBOM for audit.

---

Optional features (we recommend against enabling) ⚠️

Three optional mixins exist for per-call wrapper API, AST rewrite, and type-stub / internal API tooling. They are disabled by default and only activate when you set the corresponding environment variable:

Feature	Env var	API when enabled
Per-call wrapper API	XWLAZY_PER_CALL_API=1	lazy_import(module_name, package=..., mode=..., root=...)
AST rewrite / lazy transform	XWLAZY_AST_LAZY=1	enable_ast_lazy(root=...), disable_ast_lazy()
Type-stub / internal API tooling	XWLAZY_TYPING_TOOLS=1	attach_stub(package_name, stub_content=..., stub_path=...), get_stub_registry()

From a software engineering perspective we recommend against enabling these. They increase complexity, reduce maintainability, and (for AST/type-stub) are fragile or address a different problem domain. When enabled, they may or may not work well together (e.g. per-call and AST both touching imports). Prefer the core hook / auto_enable_lazy / attach API. Enable only for edge cases or compatibility.

---

Troubleshooting 🩺

See what’s going on:

from xwlazy import get_all_stats

stats = get_all_stats()  # installed_packages, failures, lockfile_path, keyword detection, etc.

“Nothing gets installed”: Check get_lazy_install_stats("your-package") — enabled and mode. If you use an allow list, the package must be in it.

First import slow: That’s the first install. Use full to pre-install everything, or lite and install deps yourself; caching is on by default.

---

Docs and tests 📚

Content in this README is aligned with the project REFs and docs/GUIDE_01_USAGE.md (per GUIDE_63_README).

• Start: docs/INDEX.md — doc index and quick links.
• Use it: docs/GUIDE_01_USAGE.md — modes, integration, production, troubleshooting.
• Requirements and status: docs/REF_01_REQ.md, docs/REF_22_PROJECT.md.
• API and DX: docs/REF_15_API.md, docs/REF_14_DX.md.
• Architecture: docs/REF_13_ARCH.md.
• Quality: docs/REF_54_BENCH.md, docs/REF_51_TEST.md. Benchmark run logs: docs/logs/benchmarks/.

Tests:

python tests/runner.py
# or per layer: python tests/0.core/runner.py, python tests/1.unit/runner.py

See docs/REF_51_TEST.md for test layers and coverage.

---

🔬 Positioning in the ecosystem

xwlazy combines lazy imports, on-demand installation, mapping-aware resolution, and optional audit features (lockfile, SBOM). It uses multi-tier caching (in-memory LRU + disk cache), manifest-based discovery, a deny list for problematic packages, and optional audit outputs (~/.xwlazy/xwlazy.lock.toml, ~/.xwlazy/xwlazy_sbom.toml, enabled with XWLAZY_AUDIT_ENABLED=1).

---

License and links 🔗

MIT — see LICENSE.

• Homepage: https://exonware.com
• Repository: https://github.com/exonware/xwlazy
• Contact: [e-mail] · eXonware Backend Team

Version: See version.py or PyPI. Updated: See version.py (__date__).

Built with ❤️ by eXonware.com - Revolutionizing Python Development Since 2025