# array_values should be a no-op for packed layout arrays - Thread: https://externals.io/message/98515 - Original archive: https://news-web.php.net/php.internals/98515 - Started by Jesse Schalken on 2017-03-13T05:34:40+00:00 - Messages: 7 - Reply by email to internals@lists.php.net - More for agents: search with https://externals.io/search.md?q={query} · JSON API at https://externals.io/api · documentation at https://externals.io/llms.txt --- ## #98515 — Jesse Schalken — 2017-03-13T05:34:40+00:00 URL: https://externals.io/message/98515#98515 I noticed this commit recently in HHVM, which makes `array_values()` return the same array back if it's packed instead of building a copy. My understanding is PHP 7 also has packed arrays like HHVM, and my reading of PHP's `array_values()` implementation (here ) is that it always creates a copy of the array, even if it's packed. Would this be a worthwhile optimisation to make in PHP as well? --- ## #98519 — Julien Pauli — 2017-03-14T16:21:57+00:00 URL: https://externals.io/message/98515#98519 · In reply to: #98515 On Mon, Mar 13, 2017 at 6:34 AM, Jesse Schalken wrote: > I noticed this commit > 9e853282fac968d9cd454> > recently in HHVM, which makes `array_values()` return the same array back if > it's packed instead of building a copy. > > My understanding is PHP 7 also has packed arrays like HHVM, and my reading > of PHP's `array_values()` implementation (here > 29cfb16a55196dc43/ext/standard/array.c#L4000>) > is that it always creates a copy of the array, even if it's packed. > > Would this be a worthwhile optimisation to make in PHP as well? > That would be nice if `array_values()` wouldn't reindex the array. But it does. $a = [12=> 'foo', 42 => 'bar']; // packed array $b = array_values($a); $b // [0 => 'foo', 42 => 'bar']; // keys have been changed and are not kept from source So, `array_values()` must allocate a new array, and fill it in with the values of source, because `array_values()` reindexes the source array from key 0 ; whatever was the original array. Note that the allocation is very light, in PHP 7, we allocate a full buffer at once , and not one buffer for each slot. The performance inpact of such an allocation is really really tiny. Julien.Pauli --- ## #98520 — Sara Golemon — 2017-03-14T16:48:26+00:00 URL: https://externals.io/message/98515#98520 · In reply to: #98519 > > […] > > That would be nice if `array_values()` wouldn't reindex the array. > > But it does. > > $a = [12=> 'foo', 42 => 'bar']; // packed array > $b = array_values($a); > > $b // [0 => 'foo', 42 => 'bar']; // keys have been changed and are not > kept from source > > So, `array_values()` must allocate a new array, and fill it in with the > values of source, because `array_values()` reindexes the source array from > key 0 ; whatever was the original array. > > Note that the allocation is very light, in PHP 7, we allocate a full buffer > at once , and not one buffer for each slot. The performance inpact of such > an allocation is really really tiny. > Minor nit: [12=>'foo', 42=>'bar'] is not a packed array. [0=>'foo', 2=>'bar'] is however, so your primary point stands. However it should be simple enough to detect when a packed array is also vector-like (indexed from 0 to n-1) and make this minor optimization. -Sara --- ## #98526 — Jesse Schalken — 2017-03-14T23:25:35+00:00 URL: https://externals.io/message/98515#98526 · In reply to: #98520 > > Minor nit: [12=>'foo', 42=>'bar'] is not a packed array. > > [0=>'foo', 2=>'bar'] is however, so your primary point stands. > However it should be simple enough to detect when a packed array is > also vector-like (indexed from 0 to n-1) and make this minor > optimization. > > -Sara > I didn't realise packed arrays could still have gaps. If that's the case, why can't [12=>'foo', 42=>'bar'] be packed (index 0-11 and 13-41 are undefined)? I agree checking if the array is both packed and without gaps would still make a worthwhile optimisation. --- ## #98527 — Sara Golemon — 2017-03-14T23:39:02+00:00 URL: https://externals.io/message/98515#98527 · In reply to: #98526 > I didn't realise packed arrays could still have gaps. If that's the case, > why can't [12=>'foo', 42=>'bar'] be packed (index 0-11 and 13-41 are > undefined)? > It can be, but in practice isn't likely to be since that's a lot of empty space for a two element array. It will end up mixed pretty quickly before you get to that point. > I agree checking if the array is both packed and without gaps would still > make a worthwhile optimisation. > Oh yeah, btw... six hours ago... https://github.com/php/php-src/commit/c74bc87c74f48bc55541b3bf2fc67d595f58a3b5 -Sara --- ## #98528 — Jesse Schalken — 2017-03-14T23:58:49+00:00 URL: https://externals.io/message/98515#98528 · In reply to: #98527 > > Oh yeah, btw... six hours ago... > https://github.com/php/php-src/commit/c74bc87c74f48bc55541b3bf2fc67d > 595f58a3b5 > > -Sara > Awesome, thanks Sara. --- ## #98533 — Julien Pauli — 2017-03-15T09:01:32+00:00 URL: https://externals.io/message/98515#98533 · In reply to: #98528 On Wed, Mar 15, 2017 at 12:58 AM, Jesse Schalken wrote: > > > > > […] > > Awesome, thanks Sara. > > Nice Sara. Yep my example wasn't that good , but such an array with holes could be packed, thus pretty unlikely. For interested souls , I detailed that in an article of mine, targeting 7.0 , at http://jpauli.github.io/2016/04/08/hashtables.html Julien.P