Hi everyone
I've created a PR implementing two new optimizations for closures. [1]
There are some theoretical BC breaks, so please let me know if any of
them are of concern to you.
1. Static inference
The engine will attempt to infer static for closures that are
guaranteed not to make any use of $this. This has the benefit of
avoiding a cycle when a closure is stored in a property of the
declarator object. Frequently, such objects and their closures will
not be collected for the entirety of the request, given the GC often
doesn't run at all. But the real reason for this optimization is to
enable optimization nr. 2.
It's worth noting the rules for inferring static are slightly
esoteric. You can read more about the rules in the PR description.
2. Stateless closure caching
Stateless closures, i.e. those that are static, don't capture any
variables and don't declare any static variables are now cached.
function test() {
$x = function () {};
}
for ($i = 0; $i < 10_000_000; $i++) {
test();
}
Previously, this would have created 10 000 000 closure instances, even
though all closures are effectively identical. Now, the first closure
will be kept alive for reuse. With these two optimizations, this small
benchmark improves by ~80%. Of course, this is a very synthetic
benchmark. However, I could also measure improvements in real
applications that instantiate many stateless closures. For example, in
the Laravel template these two optimizations can avoid 2384 out of
3637 closure instantiations, improving performance by ~3% on my
machine.
The foreshadowed BC breaks come down to a three things:
1. For closures that are inferred as static,
ReflectionFunction::getClosureThis() will now return NULL.
2. Objects that would previously have created cycles may be collected
earlier, also triggering destructors earlier. IMO, this is a feature
and more predictable than the current behavior.
3. Two stateless closures originating from the same lexical location
will now be identical. I.e.:
function test() {
return function () {};
}
test() === test(); // true
Of note is that Closure::bind() and Closure::bindTo() usually throw
when attempting to bind an object to a static closure. In my PR, this
is explicitly allowed only for closures that are inferred as static,
but not those that are explicitly static.
I'm looking forward to your feedback. If there are no concerns, I will
merge this PR into master for PHP 8.6 in roughly two weeks.
Ilija
[1] https://github.com/php/php-src/pull/19941
