Hi,
there are some questions whether PR #3360 (Implement ZEND_ARRAY_KEY_EXISTS
opcode to speed up array_key_exists()) [1] could go into 7.3.0beta2 or not.
This PR adds new opcode, but also significantly improves performance of
array_key_exists(), to the point that it becomes faster than isset(). This
function
is often used in major OSS libraries and some of them use hacks with
pre-checking
using isset() to avoid the performance overhead of calling this PHP
function.
Language-wise it's similar to ZEND_COUNT or ZEND_IN_ARRAY - only a specific
instruction with a VM handler to eliminate the function calling overhead.
This new opcode may have some implications for extensions, though quickly
grepping
trough XDebug and PHPDBG source didn't reveal any specific handling for
ZEND_COUNT or ZEND_IN_ARRAY.
Do you think this optimization could still go into 7.3 so the PHP ecosystem
could benefit from this improvement (possibly could be reverted in case of
issues)?
Or does it have to wait for 16 months?
Thanks,
M.
Do you think this optimization could still go into 7.3 so the PHP ecosystem
could benefit from this improvement (possibly could be reverted in case of
issues)?
My vote is no.
It's an optimization and optimizations are yay, but feature freeze
isn't just a PR label, it's a release process milestone. The tradeoff
against code-stability just isn't there.
-Sara
Do you think this optimization could still go into 7.3 so the PHP ecosystem
could benefit from this improvement (possibly could be reverted in case of
issues)?My vote is no.
It's an optimization and optimizations are yay, but feature freeze
isn't just a PR label, it's a release process milestone. The tradeoff
against code-stability just isn't there.-Sara
--
I agree with Sara. Sadly, I also have some similar improvements I
wanted to get in for improving the performance of array_slice but
couldn't meet feature freeze. If we were to bring in my changes and
yours and who knows who else may have lurking changes like this then
the stability risk Sara mentioned is real.
It hurts to wait for another release but the bright side is that is a
lot of time for the feature to get tested and be far less risky.