`--repeat` cli option

[staabm]

Description

While discussing stuff at the PHPUnit codesprint we stumbled over a comment mentioning the --repeat cli option.

running e.g. php --repeat=10 -r 'echo "hello";' yields

➜ php --repeat=10 -r 'echo "hello";'
Executing for the first time...
helloFinished execution, repeating...
helloFinished execution, repeating...
helloFinished execution, repeating...
helloFinished execution, repeating...
helloFinished execution, repeating...
helloFinished execution, repeating...
helloFinished execution, repeating...
helloFinished execution, repeating...
helloFinished execution, repeating...
hello%

it works the same way with php --repeat=10 -f myfile.php

this option was something noone of the people in the room has every seen/used.

we are wondering...

• php --help doesn't mention this option
• php --r<TAB> does not autocomplete this option

PHP Version

PHP 8.3.12 (cli) (built: Sep 24 2024 18:08:04) (NTS)

Operating System

macos 15

[bwoebi]

It's mostly an internal option for re-running tests in the face of JIT (where you want the second run to be fully JITed). Is there any use case except for run-tests to need it?

[staabm]

we discuss this php-src option while thinking about a --repeat option for PHPUnit in https://github.com/sebastianbergmann/phpunit/issues/5718.

so yeah, we think about suggesting to the PHPUnit users to use php-src php --repeat=10 phpunit mytest.php so we don't need to re-implement it in PHPUnit itself (but we don't want to advertise a php-src internal option)

[cmb69]

Has apparently been introduced by #6365; considering the long time we have it, we may make it official.

[iluuu1994]

Is there a purpose for it in userland? If we make it official, it might be impossible to improve further for internal purposes (at least without introducing more flags). E.g. it might be useful to repeat the script without clearing the pointer map. This would expose missed handling of certain conditions in optimized paths in the VM. These are often missed in tests, because they require two executions, one to fill the cache slots and another to run the optimized paths.

[bwoebi]

@staabm Is there a specific advantage specifying php --repeat=N over for n in 1..N; do php; done?

From my perspective, I'd rather include such a feature directly in phpunit, which tell would allow proper handling, like abort on first failure, giving an overview of all N runs at the end etc.?

[staabm]

We discuss this ideas and implications and expectations in https://github.com/sebastianbergmann/phpunit/issues/5718

Thanks for providing your opinions

[cmb69]

Is there a specific advantage specifying php --repeat=N over for n in 1..N; do php; done?

Performance, due to avoiding the PHP startup. An extreme case on Windows:

> type test.php
<?php
echo "hello world\n";

> hyperfine.exe "php --repeat 10 test.php" "for /l %i in (1,1,10) do @php test.php"
Benchmark 1: php --repeat 10 test.php
  Time (mean ± σ):      52.5 ms ±   2.0 ms    [User: 19.6 ms, System: 31.7 ms]
  Range (min … max):    49.6 ms …  59.1 ms    46 runs

Benchmark 2: for /l %i in (1,1,10) do @php test.php
  Time (mean ± σ):     465.0 ms ±   4.5 ms    [User: 159.1 ms, System: 300.0 ms]
  Range (min … max):   459.5 ms … 472.5 ms    10 runs

Summary
  php --repeat 10 test.php ran
    8.86 ± 0.35 times faster than for /l %i in (1,1,10) do @php test.php

[iluuu1994]

This is most noticeable for very short scripts. Most real-world applications would dwarf startup-time, it seems.