Magic with memory usage of array

[woodholly]

Description

The following code:

<?php
$a = array();
for($i=1; $i<=1000000;$i++) {
	$a[$i] = var_export(array("foobar$i"), true);
}
echo "done, memory usage = " . (memory_get_usage(true)/1024/1024) . "Mb\n";

Resulted in this output:

done, memory usage = 278.00390625Mb

But this code with var_export casted to "real" string:

<?php
$a = array();
for($i=1; $i<=1000000;$i++) {
	$a[$i] = "" . var_export(array("foob(ar$i"), true);
}
echo "done, memory usage = " . (memory_get_usage(true)/1024/1024) . "Mb\n";

Resulted in this output:

done, memory usage = 94.00390625Mb

1. Why original var_export eats x3 more memory in comparison with string-casted variant ? (same thing with json_encode)
2. Why casting using (string) does not help ?

What kind of magic is working here ?

PHP Version

PHP 7.4

Operating System

Ubuntu

[alecpl]

Interesting. I confirm this on 3v4l.org in all supported versions. Number of iterations need to be at least 10000 to notice the memory usage difference (1000000 is too big there).

[arnaud-lb]

var_export uses a string builder internally that allocates a buffer a bit larger than necessary while building the string. The issue here is that the string is not copied to a smaller buffer at the end.

The cast to string does nothing when the value is already a string. The concatenation however will force the string to be reallocated, and will use the minimum amount of memory.

[arnaud-lb]

Fixed in #8902. Thank you @woodholly !