Performance improvement of StringContains

[willemstuursma]

I've created a faster version of StringContains, in particular for large haystacks.

I noticed a 4.5x speed up for asserting 1,000x that a 2MB string contains an other string.

We don't need the multi-byte ware function mb_strpos as it is approximately O(n^2) where n is the length of the haystack.

• We don't care about the character position of the substring.
• In most encodings, the same characters will always be represented by the same bytes. This holds for ISO-8859-1 and friends, and UTF-8.

In UTF-16 and UTF-32, it is theoretically possible that same byte sequences map to different characters, but as all PHPUnit's own output is not compatible with this encoding I don't think it will cause any issues in practice as I don't think anyone is running PHPUnit with PHP's internal encoding set to UTF-16/UTF-32.

NB. There is the issue that UTF-8 allows the same characters to be represented by different combinations of unicode code points (and thus bytes), but that wasn't supported by mb_strpos anyway as it doesn't normalize the string internally.

[be-heiglandreas]

The only way to normalize strings in unicode would be to use \Normalizer::normalize($string, Normalizer::FORM_C) on both strings before passing them to the strpos-function.

But TBH I would leave that to the person writing the test to be able to find issues with different normalizations (which can be a PITA especially with filenames on MacOS-systems - been there...)

And as that will only work on unicode-strings it would require a lot of guesswork before the actual strpos regarding the current strings encoding.

Additionally while it might be possible to find a binary match in UTF16 or UTF32 when looking for a single character the possibility to find such a match decreases rapidly with the number of characters tried to match. This seems such an edge-case that the possible performance increase for the majority of the use-cases IMO verifies this risk.

[heiglandreas]

This modification would increase speed not only for binary safe comparisons but also for case-insensitive comparisons.

[heiglandreas]

To speed the case-insensitive comparison up we could use $other = mb_strtolower($other) (as used in line 45) here to convert the searched phrase to lower case which is much faster here than using mb_stripos and then do the comparison binary safe in line 84.

[sebastianbergmann]

Feel free to send a pull request for this.

[sebastianbergmann]

Thank you for your review, @heiglandreas.

[willemstuursma]

Thanks @sebastianbergmann.