phpolars

Data & Workflow

Statistics, Modeling & Visualization

Guide

Compatibility and limitations

PHP versions and platforms

  • Supported PHP versions: 8.3, 8.4, and 8.5, as composer.json allows (>=8.3 <8.6).
  • PHP build: non-thread-safe (NTS). ZTS builds are not supported.
  • Operating systems: Linux and macOS. Windows is not supported.
  • Prebuilt binaries: Linux with glibc 2.34 or later on x86_64 and arm64. Other systems build from source.

CI runs the Rust and PHP checks with PHP 8.3, 8.4, and 8.5 on Ubuntu 24.04 and macOS 15. The installer checks build each PHP version on Debian Bookworm for x86_64 and arm64, install it with PIE, and send requests through PHP-FPM.

The pipe operator |> in some examples is PHP 8.5 syntax. phpolars itself does not need it, and every example that uses it has a method-chain version for PHP 8.3 and 8.4.

Agreement with py-polars

phpolars uses the Polars 0.55.2 Rust crates. The test suite compares phpolars with py-polars 1.43.2, the release whose Rust core is closest to 0.55.2. scripts/pyfixtures.py builds 314 cases with py-polars, and tests/pyfixtures.php builds the same cases with phpolars. A case passes when the column names, the data types, and every value agree. Floats may differ by a relative tolerance of 1e-9. Cases that expect an error pass when phpolars raises the same exception class. All 314 cases pass.

The cases cover data frame methods, functions, joins, grouping, windows, sorting, casts, SQL, and the str, dt, list, struct, and arr namespaces. A method or argument outside those cases follows py-polars’ names, but its agreement with py-polars has not been tested. The verification page lists every case.

Not available

  • Python callables. py-polars methods that run a Python function per value or per batch, such as map_elements, map_batches, fold, and map_groups, have no equivalent, because PHP code cannot run on Polars threads. DataFrame::mapRows() runs a PHP callable per row on the PHP thread, after collecting.
  • The GPU engine. collect(engine: 'gpu') throws a ValueError.
  • Conversion to pandas, NumPy, or PyArrow objects. Data leaves phpolars as PHP values, as files, or through PDO.
  • Formats left out of a build. Parquet, IPC, JSON, SQL, and table formatting are Cargo features, and an extension built without one raises a ComputeError from that format’s functions. The release binaries include all of them.

From phpstats/phpolars@8487062.