Streamlined the mechanics behind composed properties. #340
Conversation
andersio
force-pushed
the
property-refactor
branch
11 times, most recently
from
5b4b2ce
to
a03fdf3
Compare
andersio
force-pushed
the
property-refactor
branch
4 times, most recently
from
6408cd9
to
d1c58ff
Compare
Sources/Property.swift
Outdated
| init(_ value: Value) { | ||
| self.value = value | ||
| } | ||
| } |
There was a problem hiding this comment.
I'd like to see this use the same approach as Atomic to prevent unsafe accesses—hiding value and lock behind locking APIs.
Sources/Property.swift
Outdated
| } | ||
|
|
||
| extension NSRecursiveLock { | ||
| fileprivate func sync<Result>(_ action: () throws -> Result) rethrows -> Result { |
There was a problem hiding this comment.
I think this can go away of appropriate methods are added to PropertyBox.
Sources/Property.swift
Outdated
| fileprivate func sync<Result>(_ action: () throws -> Result) rethrows -> Result { | ||
| lock(); defer { unlock() } | ||
| return try action() | ||
| func modifyDidSet<Result>(_ action: (inout Value) throws -> Result, didSet: (() -> Void)?) rethrows -> Result { |
There was a problem hiding this comment.
Just make this:
func modify<Result>(
_ action: (inout Value) throws -> Result,
didSet (() -> Void)? = nil
) rethrows -> Result {?
There was a problem hiding this comment.
Ah, I thought the trailing closure would be in conflict. But I just realize didSet can be moved as the first. 😛
Sources/Property.swift
Outdated
| } | ||
| }, didSet: { | ||
| observer.action(event) |
There was a problem hiding this comment.
Can't observer.action be in the action block? It still happens inside the lock either way.
There was a problem hiding this comment.
It would violate access exclusivity because of the "de jure" copy-in, copy-out semantic of inout being in conflict with the guarantee of property observers always seeing the latest value.
There was a problem hiding this comment.
Ah, MutableProperty.modify needs to be changed too.
andersio
force-pushed
the
property-refactor
branch
from
d27690e
to
4510c3b
Compare
andersio
force-pushed
the
property-refactor
branch
7 times, most recently
from
b395238
to
af6a7f6
Compare
andersio
force-pushed
the
property-refactor
branch
from
af6a7f6
to
35e1202
Compare
Sources/Atomic.swift
Outdated
| // default. Since fairness is not a concern of our reactive primitives | ||
| // anyway, it can be disabled. | ||
| pthread_mutexattr_setpolicy_np(attr, _PTHREAD_MUTEX_POLICY_FIRSTFIT) | ||
| #endif |
There was a problem hiding this comment.
I don't love setting an untested (in the real world) policy for only older Apple OSes.
Because this would affect a small portion of devices, it's hard to have confidence that it works well. I'd rather see us take a performance hit on older OS versions and not have a separate code path that's hard to test.
Replaced
replayLazily(upTo:)with a custom implementation.Composed properties now cache only one copy of the latest value, down from 1-2 copies currently. Moreover, it now stops caching after both the
Propertyand all the producer instances has deinitialised.