Lazy Evaluation in PHP — Generators & Iterators Demo

This repository accompanies the article:
Lazy Evaluation in PHP: How Generators and Iterators Save Memory and Speed up Code
— dev.to: https://dev.to/oleksandr_vasyliev/lazy-evaluation-in-php-how-generators-and-iterators-save-memory-and-speed-up-code-3529

It demonstrates the difference in memory usage and runtime when processing large datasets with:

• classic arrays (eager loading)
• generators / iterators (lazy loading)

Author: Oleksandr Vasyliev

Requirements

• PHP 8.2+ (works with 8.1+, but examples target 8.2)
• Composer (optional, for running PHPUnit tests)

Quick start

# 1) Clone your fork, then:
cd php-lazy-evaluation-demo

# 2) Install dependencies
composer install 

# 3) Generate sample CSV & NDJSON data (100k rows by default)
php bin/make_sample.php --rows=100000

# 3) Run the CSV demo
php examples/array_vs_generator.php            # compare both
php examples/array_vs_generator.php array      # only eager
php examples/array_vs_generator.php generator  # only lazy

# 5) Run the API (NDJSON) streaming demo
php examples/batch_api_processing.php

Benchmark (1,000,001 CSV rows)

Numbers from a Windows + OSPanel run (your mileage may vary):

Method	Time	Memory used	Peak diff	Rows
Array (eager)	1.401s	120 B	395.92 MB	1,000,001
Generator	1.012s	0 B	0.00 MB	1,000,001

These results show that the eager approach spikes peak memory (loading the whole dataset), while the generator keeps memory flat by streaming one row at a time.

Reproducing your own numbers

• Increase the dataset size:

  php bin/make_sample.php --rows=5000000

• Run array and generator separately to isolate their memory profiles:

  php examples/array_vs_generator.php array
  php examples/array_vs_generator.php generator

• Optionally redirect output to a log:

  php examples/array_vs_generator.php > results/benchmark.txt

Note: In the NDJSON benchmark, both eager and lazy modes show similar peak memory. This is because each decoded JSON object is still built fully into memory and processed in-place. Lazy reading helps mostly when:

• the source format is line-based or record-based (CSV, line-JSON),
• each record is processed incrementally and not accumulated.

In contrast, the CSV benchmark clearly benefits from lazy iteration, especially with millions of rows.

Visual Comparison

Tests

php ./vendor/bin/phpunit

License

MIT