[Console] avoid using huge amount of memory when formatting long exception

[paxal]

Q	A
Branch?	master
Bug fix?	no
New feature?	no
BC breaks?	no
Deprecations?	no
Tests pass?	yes
Fixed tickets	none
License	MIT

When formatting exceptions, preg_split('//u') is used to iterate over utf8 characters. When the exception is long, the amount of memory used is huge and can reach memory_limit.

This PR uses a Generator to iterate over the string instead of splitting it, thus reducing the amount of memory.

[nicolas-grekas]

Can you give us some numbers you measured about this improvement?

[paxal]

Please don't judge me ‒ my exception was 1166286 (~1 MB) chars long (very very long doctrine request).

Snippet :

<?php

use Symfony\Component\Console\Application;
use Symfony\Component\Console\Output\NullOutput;

require_once __DIR__.'/vendor/autoload.php';

$start = microtime(true);

$application = new Application();
$longText = str_repeat('Oh my god ! ', 100000);
$application->renderException(new \Exception($longText), new NullOutput());
echo sprintf('%d ms / %d MB', 1000*(microtime(true)-$start), memory_get_peak_usage()/1024**2).PHP_EOL;

Results before (faster, but a lot of memory) : 963 ms / 106 MB
Results after (longer, but less memory) : 5337 ms / 8 MB
With a chunk size of 10 000 (seems good enough to me, I'll send a small patch) : 964 ms / 8 MB

[nicolas-grekas]

Thank :)
Can you please try this patch instead?
Does it perform better?

--- a/src/Symfony/Component/Console/Application.php
+++ b/src/Symfony/Component/Console/Application.php
@@ -1149,7 +1149,11 @@ class Application implements ResetInterface
         $utf8String = mb_convert_encoding($string, 'utf8', $encoding);
         $lines = [];
         $line = '';
-        foreach (preg_split('//u', $utf8String) as $char) {
+        $offset = 0;
+        while (preg_match('/.{1,1000}/u', $utf8String, $m, 0, $offset)) {
+            $offset += \strlen($m[0]);
+
+            foreach (preg_split('//u', $m[0]) as $char) {
             // test if $char could be appended to current line
             if (mb_strwidth($line.$char, 'utf8') <= $width) {
                 $line .= $char;
@@ -1158,6 +1162,7 @@ class Application implements ResetInterface
             // if not, push current line to array and make new line
             $lines[] = str_pad($line, $width);
             $line = $char;
+            }
         }

[paxal]

Tested with 1000 : 1337 ms / 8 MB
Tested with 10000 : 1002 ms / 8 MB

Which value do you want me to apply for the patch ?

[nicolas-grekas]

10k is good

[paxal]

I think we're good 👍 Thanks @nicolas-grekas

[xabbuh]

For 4.4?

[chalasr]

Thank you @paxal.