phplyr
Benchmarks
Timings from the benchmark files in the phplyr repository. Each report gives the machine and date it was recorded, so an older report may not reflect the current code.
phplyr backends and dplyr
Three pipelines on a generated sales table of 100,000 and 1,000,000 rows, joined to a 100-row store table. phplyr runs each on phibble's in-memory DataFrame, on PolarsTable (Polars at 2 threads), and on SQLite through SqlTable. dplyr runs the first two pipelines on the same data. The pipeline with a PHP closure has no dplyr version. Every backend returned the same rows as DataFrame, and dplyr's row counts and column sums matched.
Apple M5 Ultra · PHP 8.5.10 · recorded 2026-10-18 · commit ed999d8
| Case | phplyr on DataFrame | phplyr on PolarsTable | phplyr on SQLite | dplyr 1.2.1 |
|---|---|---|---|---|
| Filter, join, summarise, and rank (100,000 rows) | 0.089 | 0.006 | 0.043 | 0.033 |
| Filter, join, summarise, and rank (1,000,000 rows) | 0.751 | 0.051 | 0.434 | 0.065 |
| Grouped windows over every row (100,000 rows) | 0.095 | 0.043 | 0.311 | 0.004 |
| Grouped windows over every row (1,000,000 rows) | 1.25 | 0.415 | 3.33 | 0.020 |
| Filter with a PHP closure, join, summarise, and rank (100,000 rows) | 0.116 | 0.078 | 0.197 | |
| Filter with a PHP closure, join, summarise, and rank (1,000,000 rows) | 1.13 | 0.874 | 2.01 |
Best of 3 runs; pipeline through collect(); data preloaded and not timed, in s. Values without a measurement are left out of the chart.