Modifying length of file can break VCR integration

[lkysow]

Tl;dr: Changing the length of your file by adding/removing code can break PHP VCR.

How The Code Works

PHP VCR modifies files when they're being read from the filesystem through the use of PHP's php_user_filter (http://php.net/manual/en/class.php-user-filter.php).

When a file is read from the filesystem, it goes through this function in PHP VCR

<?php
public function filter($in, $out, &$consumed, $closing)
{
        while ($bucket = stream_bucket_make_writeable($in)) {
            $bucket->data = $this->transformCode($bucket->data);
            $consumed += $bucket->datalen;
            stream_bucket_append($out, $bucket);
        }

        return PSFS_PASS_ON;
    }
}

which calls transformCode($code) in each code transformer

<?php

class CurlCodeTransform extends AbstractCodeTransform
{

    private static $patterns = array(
        '/(?<!::|->)\\\?curl_init\s*\(/i' => '\VCR\LibraryHooks\CurlHook::curl_init('
        ...
    );

    protected function transformCode($code)
    {
        return preg_replace(array_keys(self::$patterns), array_values(self::$patterns), $code);
    }
}

The transformCode($code) function does a preg_replace and replaces the original code with code that calls PHP VCR.

The Problem

You may have noticed that the filter function operates in a while loop calling stream_bucket_make_writeable. The bug is that this function can split up the stream, i.e. the file that is being read.

If it splits it between a line of code that we're trying to search and replace, then the preg_replace will fail and we won't modify this file.

Example

The class we're rewriting:

<?php
class TransformMe {
    public function test() {
        $curl = curl_init();
    }
}

Ideally, this will be re-written to

<?php
class TransformMe {
    public function test() {
        $curl = \VCR\LibraryHooks\CurlHook::curl_init();
    }
}

However if the file is split like this:

<?php
class TransformMe {
    public function test() {
        $curl = curl_

init();
    }
}

Then the transformation will never happen.

This was fun to debug :)

Solutions?

I'd say the best thing to do would be to read the entire file at once, and then do the search and replace, rather than relying on PHP's streams and bucketing.

[adri]

Great analysis @lkysow. Thank you very much!
I agree, it should read the entire file before replacing data. I'll implement it asap.

[adri]

Before merging #77 I would like to write a regression test. Do you have an example where it failed?
Otherwise I'll make something up ;-)

[lkysow]

@adri thanks for the fast response! I've managed to reproduce it with this file: https://gist.github.com/lkysow/518afbebcaaefd486024