Greg Schofield

This commit performs a significant **dependency upgrade**, updating the `uv` package manager from version 0.9.22 to **0.10.11** across the project. This **maintenance** effort primarily impacts the **`packages/python` module**, its associated **CI/CD workflows** for releases and tests, and the `vercel-runtime` dependency. A minor **observability enhancement** is included, as the `build` function in `packages/python/src/index.ts` now logs the exact `uv` version using `execSync`. Test fixtures are also updated to align with the new `uv` version's logging format, ensuring continued compatibility and reflecting potential breaking changes.

This commit implements a **bug fix** within the **Python Lambda bundler** to resolve an issue where deployments would erroneously fail if the bundled artifact was exactly 245MB. Previously, the **Lambda bundle size validation** incorrectly reported bundles at the precise 245MB limit as exceeding it, preventing successful deployment. The change introduces a **100KB buffer** to the final size verification logic in `packages/python/src/dependency-externalizer.ts`, specifically within the `externalize` function. This ensures that **Python Lambda deployments** can proceed successfully when the bundle size is exactly at the 245MB threshold, significantly improving the reliability of the deployment process for large functions.

This commit **optimizes resource utilization** for **Python runtime dependency installations** by reverting to the use of hardlinks. Specifically, it **refactors** the `vercel-runtime` module, modifying the `uv sync` command in `vc_init.py` to explicitly use `--link-mode hardlink` instead of `copy`. This change significantly **reduces peak disk space consumption** in temporary directories (e.g., `/tmp`), improving efficiency during dependency setup. By leveraging hardlinks, the system avoids unnecessary file duplication, leading to a more streamlined and less resource-intensive installation process for **Python deployments**. A new changeset entry also documents this **performance improvement**.

This commit introduces a **bug fix** to the **Python dependency externalizer** by adjusting the `LAMBDA_SIZE_THRESHOLD_BYTES` constant. Previously, this threshold did not account for the size of Lambda layers, causing deployment failures for functions nearing the 250MB limit. The threshold is now **reduced to 245MB**, providing a crucial 5MB buffer for layers and ensuring the `remainingCapacity` calculation within `externalizeDependencies` accurately reflects the available space. This **maintenance** update prevents unexpected build failures and improves the reliability of **Python Lambda function deployments** by ensuring more accurate size checks.

This commit performs a **maintenance adjustment** to the **Python Lambda packaging process**, specifically by **decreasing the target size for dependency externalization**. The `LAMBDA_PACKING_TARGET_BYTES` constant in `packages/python/src/dependency-externalizer.ts` is reduced from 245MB to 240MB. This change provides **more buffer space for user-specific source code** within the overall Lambda package, aiming to prevent deployment issues related to package size limits. The update includes corresponding **test modifications** and **documentation** for this configuration change.

This commit **reverts** the previous implementation that introduced caching for Python virtual environments (`.venv`) and `uv` caches within the **Python build system** (`packages/python`). This **revert** acts as a **fix** for critical issues where the caching mechanism caused `.venv` directories to grow excessively with stale packages, leading to inflated cache sizes and incorrect runtime uncompressed size calculations for Python functions. It systematically removes the `prepareCache` function and all associated `uvCacheDir` logic from the build process, installation routines, and utility functions like `ensureVenv` and `createVenvEnv`. This change ensures accurate dependency externalization and prevents unintended cache bloat, thereby improving the reliability of function size reporting.

This commit **fixes a bug** in the **`python-analysis`** package by improving the parsing of complex **Python environment markers** in `requirements.txt` files. It specifically addresses issues with expressions combining `and` and `or` logic within parentheses. This is achieved by **updating the `pip-requirements-js` dependency** to version `1.0.3`, which contains the necessary parsing enhancements. Additionally, **new test cases** have been added to `packages/python-analysis/test/requirements-txt-parser.test.ts` to ensure robust and accurate handling of nested environment marker parsing. This change significantly enhances the reliability of Python dependency resolution within the project.

This commit introduces a **new capability** to the **Python builder** by implementing **caching for virtual environments** (`.venv`) and the `uv` cache directory. It leverages the `prepareCache` build output API to store these artifacts, which will significantly improve build performance for Python projects. The `UV_CACHE_DIR` is now consistently managed and located within the `.vercel` directory, requiring **refactoring** across several modules including `install.ts`, `utils.ts`, and `uv.ts` to properly integrate and utilize this new cache location. This change streamlines the **Python build process** and development environment setup by reducing redundant installations.

This commit introduces **observability enhancements** to the **Python builder** by integrating Datadog tracing. It **adds new capabilities** to monitor the build process, wrapping critical operations like version detection, virtual environment creation, installation, compilation, and bundling in dedicated spans within `packages/python/src/index.ts`. Furthermore, custom span tags are added to expose key metrics such as `python.version`, `python.bundle.totalSizeBytes`, and `python.bundle.runtimeInstallEnabled`, significantly improving the ability to debug and analyze Python deployments. The `DependencyAnalysis` interface in `packages/python/src/dependency-externalizer.ts` was also extended to include `totalBundleSize` for these traces. This instrumentation provides deeper insights into the performance and behavior of Python builds.

This commit **refactors** the management of the `MPLCONFIGDIR` environment variable for **Matplotlib** within the **Python runtime initialization**. It **moves** the setting from the `vc_init.py` script to a new, dedicated **TypeScript-based quirks system**. A new quirk in `packages/python/src/quirks/matplotlib.ts` now explicitly sets `MPLCONFIGDIR` to `/tmp`, centralizing this configuration. This change improves the organization and consistency of **environment variable management** for Python deployments, with **test coverage** added to validate the new quirk's behavior.

This commit **fixes a bug** in the **Python lambda build process** where projects using local, editable build dependencies would fail runtime dependency installation pre-deploy checks. It introduces a **new `noInstallProject` option** to the `UvRunner`'s `sync` method, which passes the `--no-install-project` flag to the underlying `uv sync` command. This **new capability** allows the build system to correctly handle these dependencies, preventing deployment failures for **large lambdas** that include private packages. The change primarily affects the `packages/python` module, ensuring smoother deployments for a specific class of Python projects.

This commit introduces a **defensive validation** for **Python projects** within the build system, specifically affecting **dependency externalization**. It **disables runtime dependency installations** for functions that utilize custom build or install commands, preventing potential conflicts or unexpected behavior. Furthermore, it now throws a helpful `NowBuildError` if such a function exceeds the 250MB bundle size limit, providing clear guidance to users. This **feature enhancement** ensures more robust deployments and better error reporting for complex Python build configurations.

This commit introduces a **defensive check** within the **`@vercel/python` package** to improve the robustness of its **dependency externalization and vendoring process**. Specifically, it **fixes** an issue where the system would attempt to mirror files listed in `.dist-info/RECORD` even if they were missing from the disk, potentially causing errors. By adding an `fs.existsSync` check within the `mirrorPackagesIntoVendor` function in `packages/python/src/dependency-externalizer.ts`, the system now **gracefully skips missing files**, preventing build failures and ensuring more reliable package vendoring. This change enhances the stability of Python deployments by making the vendoring logic more resilient to incomplete or malformed package structures.

This commit introduces a major **refactoring** and **performance optimization** for the **Python Lambda runtime**, specifically targeting cold-start times. It implements a new **`PythonDependencyExternalizer`** class that utilizes a **knapsack algorithm** to intelligently bundle dependencies for optimal packing. The runtime installation process within the **`@vercel/python` builder** and **Lambda runtime initialization** is updated from `uv pip install` to **`uv sync --inexact --frozen`**, driven by a new JSON configuration. This comprehensive change significantly **improves cold-start performance** for Python functions, especially those with oversized dependency bundles, by streamlining how dependencies are packaged and installed.

This commit **fixes a regression** in the **Python build system** that prevented the `VERCEL_PYTHON_ON_HIVE` environment variable from correctly skipping runtime dependency installation. It introduces a new function, `shouldEnableRuntimeInstall`, within `packages/python/src/index.ts` to conditionally bypass this process when running on the **Hive platform**. This ensures that **Python deployments utilizing Hive** will correctly optimize their build steps by avoiding unnecessary dependency installations, restoring expected behavior and improving efficiency for affected users. The change is thoroughly covered by new unit tests and documented in a changeset.

This commit introduces a **new feature** for the **Python Lambda runtime**, enabling **larger Python functions (up to 512MB) by default**. It **removes the `VERCEL_EXPERIMENTAL_PYTHON_UV_INSTALL_ON_STARTUP` experimental flag**, making runtime dependency installation a standard behavior for Python deployments. This change significantly improves the **build and deployment process** for Python serverless functions, allowing developers to deploy projects with more extensive dependency trees without manual configuration. The update also defines and tests the `LAMBDA_EPHEMERAL_STORAGE_BYTES` constant to manage ephemeral storage limits.

This commit introduces a **new capability** to support **Python functions exceeding the 250MB AWS Lambda size limit** by enabling runtime dependency installation. It enhances the **`@vercel/python` builder** to package the `uv` binary and install public dependencies on startup, while ensuring private packages are always bundled directly. The new **`@vercel/python-analysis`** module provides crucial logic for parsing `uv.lock` files and classifying packages. This feature, activated via an experimental environment variable, significantly expands the scale of Python applications deployable on the platform.

This commit introduces a **new capability** to the **Python runtime** for **Vercel Lambda functions**, enabling the installation of runtime dependencies during cold starts. It modifies the **Lambda initialization process** within `vercel_runtime/vc_init.py` to execute `uv pip install` for packages listed in `_runtime_requirements.txt`. Furthermore, this **infrastructure change** dynamically alters `sys.path` to prioritize these newly installed dependencies by inserting `/tmp/_vc_deps_overflow` at the front. This is a **high-risk change** that significantly impacts the **cold start behavior** and dependency resolution for Python serverless functions.

This commit introduces a **bug fix** to the **Python build process** by refining the application of the UV Python prefix filter. It modifies the `getAvailablePythonVersions` function within `packages/python/src/uv.ts` to ensure this specific Python version filter is **only applied when builds are executed within the Vercel infrastructure**. This change **fixes broken local `vercel build` commands for Python projects**, which were previously failing due to the filter being incorrectly applied outside the Vercel environment. The update **restores expected functionality for local development workflows** involving Python.

This commit **fixes a critical bug** in the **Python environment setup** that caused an incompatibility when a `.python-version` file specified Python 3.13 or 3.14 without an existing `pyproject.toml`. Previously, the system would incorrectly generate a `pyproject.toml` with an older Python version constraint, leading to conflicts. The **fix** modifies the `build` function in `packages/python/src/index.ts` and updates `createPyprojectToml` and `ensureUvProject` in `packages/python/src/install.ts` to correctly utilize the detected Python version for `pyproject.toml` generation. This **enhances compatibility** for users adopting newer Python versions, ensuring a smooth project setup experience within the **Python package management system**.