-
-
Notifications
You must be signed in to change notification settings - Fork 654
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
flow: Annotate some more exported React components. #4914
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks @chrisbobbe ! Generally these look good; comments on some of them below.
src/animation/AnimatedComponent.js
Outdated
|
||
const prevVisible = usePrevious(visible); | ||
|
||
const animatedValue = useRef(new Animated.Value(0)); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmm, a bit awkward that this means we're creating a new Animated.Value
on each subsequent render and then throwing it away.
But I don't see a good way in the Hooks API to avoid that. Hopefully that constructor isn't expensive 🤷
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmm, yeah, interesting...
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
See "You might also occasionally want to avoid re-creating the useRef() initial value" here. The suggested solution there looks a little awkward though.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Cool, yeah. That looks like something we'd want to do if we actually know the constructor is expensive, but not something we'd want to default to speculatively.
src/animation/AnimatedComponent.js
Outdated
const animate = useCallback(() => { | ||
Animated.timing(animatedValue.current, { | ||
toValue: visible ? fullValue : 0, | ||
delay, | ||
duration: 300, | ||
useNativeDriver: this.props.useNativeDriver, | ||
useNativeDriver, | ||
easing: Easing.out(Easing.poly(4)), | ||
}).start(); | ||
} | ||
}, [delay, fullValue, useNativeDriver, visible]); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Huh, I'm surprised the linter isn't making you mention animatedValue
in the dependencies list.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It's not clear in the docs, but I think this turns out to be expected:
facebook/react#16121 (comment)
facebook/react#14387 (comment)
facebook/react#14387 (comment)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmm, those comments explain that we shouldn't put animatedValue.current
as a dependency, and I think I understand that.
But what I'd expect the linter to look at in the first place is the local variables that are getting closed over, so animatedValue
, not any of its properties. In the first linked comment's terms, that's a ref container, not an example of ref.current
.
It's actually just fine to not have it in the dependencies list, because it comes straight from a useRef
, and that's guaranteed to give the same value on re-render. So having it as a dependency would never have any effect, because it'll never see a change.
I just wouldn't have expected the linter to detect that nuance, so I'd have expected it to flag this.
... Oh right, but this linter is from the React developers -- it's not something a random person decided to write. So maybe it's not so surprising that they did the work to capture that. And in fact it looks like this (and the other mention of "useRef" a bit lower down) is the pertinent bit of logic. The whole file is 1814 lines long -- so yeah, they weren't slacking.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Interesting; yeah, it's nice to see that they took care writing that lint rule.
src/animation/AnimatedComponent.js
Outdated
useLayoutEffect(() => { | ||
if (prevVisible !== visible) { | ||
animate(); | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Use `useLayoutEffect` [3] instead of `useEffect` because it seems
intuitively what we'd want, for starting the animation.
I think this isn't a case where we have a reason to make an exception to the general upstream advice of preferring useEffect
.
The thing that useLayoutEffect
accomplishes relative to useEffect
is that if it updates something, that update takes effect before the next paint.
Here, we're starting an animation. But the thing about an animation is that it begins at zero -- at the instant you start it, it doesn't yet have an effect, and its effects come in gradually. So I think synchronously starting the animation shouldn't have any effect on the current frame. We can let this frame go ahead and paint without trying to hold it up, and then start the animation so it happens over the next few frames.
src/animation/AnimatedComponent.js
Outdated
useLayoutEffect(() => { | ||
if (prevVisible !== visible) { | ||
animate(); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
AnimatedComponent: Only call `animate` on visibility change.
Instead of on every render, where it often ends up being unnecessary
work.
Hmm, I think there is one behavior change here that may not be desirable:
In the old code, if when this component first gets mounted it has visible: true
, it will have the value start at 0 and animate up to fullValue
.
In the new code, if when this component first gets mounted it has visible: true
... hmm, in fact this seems definitely not desirable: I think what will happen is that the value will start at 0 and will stay there, until/unless visible
changes to false and then to true again.
One way to look at this is that what we really want here is a usePrevious
that doesn't have the funny quirk where on the first call, it takes the "previous" value to be the current one. Instead, it should take another argument to explicitly tell it the "initial" value, and pass that initial value to its useRef
.
Then here, we're using the value 0 when we initialize animatedValue
; and so for initializing prevVisible
we should use the value of visible
that corresponds to that, namely false.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
In the old code, if when this component first gets mounted it has
visible: true
, it will have the value start at 0 and animate up tofullValue
.
Hmm, interesting. Is this what we want, though? If visible
is initially true, might we want to initialize animatedValue.current
with new Animated.Value(fullValue)
instead of new Animated.Value(0)
, so it would start as visible, instead of effectively starting as invisible (but with an immediately scheduled animation to become visible)?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah, that does seem better.
Looking at existing uses of this component, there's just one: it's in ComposeMenu
, and it always starts with visible
false. So we're currently not exercising this case at all. That's good because it means we don't have anything relying somehow on the current kind-of-quirky behavior.
src/common/InputWithClearButton.js
Outdated
this.setState({ | ||
canBeCleared: text.length > 0, | ||
text, | ||
}); | ||
if (this.props.onChangeText) { | ||
this.props.onChangeText(text); | ||
} | ||
}; | ||
const handleChangeText = useCallback( | ||
(_text: string) => { | ||
setCanBeCleared(_text.length > 0); | ||
setText(_text); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmm, so this is a case where we have two state fields that are tied closely together and really should be updated synchronously with each other. Does useState
guarantee that if you call two different setState functions one right after the other, nothing else will see a state where just one update has happened? I don't see anything about that in the docs:
https://reactjs.org/docs/hooks-reference.html#usestate
It may be cleanest to make a practice of keeping any such group of state fields together in an object which gets maintained as a single useState
state.
In this particular case… it sure looks like canBeCleared
is just always the same as text.length > 0
. So the best solution here might be a prep commit that just drops canBeCleared
from the state, and computes it where we use it.
src/common/SmartUrlInput.js
Outdated
}; | ||
}, []); | ||
|
||
const renderPlaceholderPart = useCallback( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe skip the useCallback
on this one. It's not any kind of event-handler callback -- rather it's a helper for rendering, to render a fragment of the element tree. And both the name, and the contents (consisting mostly of JSX, so producing some React elements to return), make it look very much like that's going to be the case.
useFocusEffect( | ||
useCallback(() => { | ||
if (textInputRef.current) { | ||
// `.current` is not type-checked; see definition. | ||
this.textInputRef.current.focus(); | ||
textInputRef.current.focus(); | ||
} | ||
}); | ||
} | ||
|
||
componentWillUnmount() { | ||
if (this.unsubscribeFocusListener) { | ||
this.unsubscribeFocusListener(); | ||
} | ||
} | ||
|
||
handleChange = (value: string) => { | ||
this.setState({ value }); | ||
}, []), | ||
); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I didn't find the doc super clear about exactly how useFocusEffect
behaves, so I looked at the implementation. It appears to be in node_modules/@react-navigation/core/src/useFocusEffect.tsx
.
There's one thing it does which the old code here didn't, but I think has no practical effect for a different reason: if the screen is already focused when the component first mounts, then useFocusEffect
will call the callback, while the old code relying on just a listener wouldn't.
(The doc does mention this, lower down.)
Empirically, when I hit the "Add new account" button to open the screen this appears on, the input gets focused anyway (before these changes). I guess that's due to the autoFocus
prop we pass to the TextInput
.
I think I wouldn't call this commit NFC, because it does take some more investigation (like seeing that autoFocus
prop) to see that it doesn't end up changing how the app visibly behaves.
Use `useRef` for `animatedValue` [1] [2]. Greg points out that we should choose `useEffect` over `useLayoutEffect` [3]. [1] https://reactjs.org/docs/hooks-faq.html#is-there-something-like-instance-variables [2] https://reactnative.dev/docs/animated [3] zulip#4914 (comment)
Greg points out [1] that `canBeCleared` is always the same as `text.length > 0`. So, just drop this piece of state and compute the value we need when we need it. [1] zulip#4914 (comment)
Following the doc at https://reactnavigation.org/docs/use-focus-effect/. Greg points out [1] that `useFocusEffect` seems to have the effect of calling the callback even on the component's first mount (if the screen is focused), whereas the old listener-based code would not. That doesn't seem to have a practical effect, here; empirically, it looks like the input gets focused anyway before this change. That's probably due to the `autoFocus` prop we pass to the `TextInput`. [1] zulip#4914 (comment).
1cc55eb
to
f18ca43
Compare
Thanks for the review! Revision pushed. Please also see my replies above: |
These are equivalent, but Greg prefers this form because it seems cleaner. See https://chat.zulip.org/#narrow/stream/243-mobile-team/topic/Flow.20types-first/near/1239060.
Like we did in 4aedc4c.
Like we did in 4aedc4c.
Like we did in 4aedc4c.
Use `useRef` for `animatedValue` [1] [2]. Greg points out that we should choose `useEffect` over `useLayoutEffect` [3]. [1] https://reactjs.org/docs/hooks-faq.html#is-there-something-like-instance-variables [2] https://reactnative.dev/docs/animated [3] zulip#4914 (comment)
Greg points out [1] that `canBeCleared` is always the same as `text.length > 0`. So, just drop this piece of state and compute the value we need when we need it. [1] zulip#4914 (comment)
Following the doc at https://reactnavigation.org/docs/use-focus-effect/. Greg points out [1] that `useFocusEffect` seems to have the effect of calling the callback even on the component's first mount (if the screen is focused), whereas the old listener-based code would not. That doesn't seem to have a practical effect, here; empirically, it looks like the input gets focused anyway before this change. That's probably due to the `autoFocus` prop we pass to the `TextInput`. [1] zulip#4914 (comment).
f18ca43
to
43a5b55
Compare
render(): React$Node { | ||
render(): Node { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
flow: Use `Node` type exported from React, instead of `React$Node`.
111 files changed, 246 insertions(+), 156 deletions(-)
Eep -- I hope you found a way to automate this!
(I would probably do something like this with a command involving perl -i -0pe
.)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I did a VSCode find-and-replace-all :)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
(More details in chat: https://chat.zulip.org/#narrow/stream/243-mobile-team/topic/codebase-wide.20edits/near/1240343)
src/animation/AnimatedComponent.js
Outdated
|
||
const prevVisible = usePrevious(visible); | ||
|
||
const animatedValue = useRef(new Animated.Value(0)); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Cool, yeah. That looks like something we'd want to do if we actually know the constructor is expensive, but not something we'd want to default to speculatively.
Thanks for the revision! All looks good now modulo the AnimatedComponent issue (#4914 (comment)). I also replied on each of those comment threads you linked to. |
Currently, AnimatedComponent has only one caller, and it always starts by passing `false` for `visible`. Since it never passes `true` initially, the new path isn't exercised. But we should make it so future callers that pass `true` initially will actually see the element start as visible. So, do that. See more discussion at zulip#4914 (comment).
Greg points out [1] that `canBeCleared` is always the same as `text.length > 0`. So, just drop this piece of state and compute the value we need when we need it. [1] zulip#4914 (comment)
Following the doc at https://reactnavigation.org/docs/use-focus-effect/. Greg points out [1] that `useFocusEffect` seems to have the effect of calling the callback even on the component's first mount (if the screen is focused), whereas the old listener-based code would not. That doesn't seem to have a practical effect, here; empirically, it looks like the input gets focused anyway before this change. That's probably due to the `autoFocus` prop we pass to the `TextInput`. [1] zulip#4914 (comment).
43a5b55
to
b184f62
Compare
Thanks for the review! Revision pushed. |
It's intentional that both of these expressions do the same thing: the point is that when the component first renders, it should start out in the desired state, and then when the desired state changes and causes a rerender it'll animate to the new desired state. So, we can make that explicit (and avoid some future change causing an accidental discrepancy) by writing the expression just once. Also because `visible` is already boolean, we can skip the `=== true`.
Instead of on every render, where it often ends up being unnecessary work.
Using `useRef` for `unsubscribeFocusListener` [1]. We'll simplify the navigation-focus logic with `useFocusEffect` in the next commit. [1] https://reactjs.org/docs/hooks-faq.html#is-there-something-like-instance-variables
Following the doc at https://reactnavigation.org/docs/use-focus-effect/. Greg points out [1] that `useFocusEffect` seems to have the effect of calling the callback even on the component's first mount (if the screen is focused), whereas the old listener-based code would not. That doesn't seem to have a practical effect, here; empirically, it looks like the input gets focused anyway before this change. That's probably due to the `autoFocus` prop we pass to the `TextInput`. [1] zulip#4914 (comment).
This is much easier to do now that we've converted these to function components in this series. This will help move us along toward Flow's new "Types-First" mode; switching entirely is zulip#4907. Prefer the `Node` type exported from React over `React$Node`. They're equivalent, but Greg prefers `Node` as it seems cleaner; see https://chat.zulip.org/#narrow/stream/243-mobile-team/topic/Flow.20types-first/near/1239060.
b184f62
to
81a50bb
Compare
Thanks, looks good! Merging. I added one small commit further tweaking that AnimatedComponent code: |
Thanks for the review!
Looks great, thanks! |
As Greg pointed out a few weeks ago [1], `usePrevious` has had a "funny quirk where on the first call, it takes the "previous" value to be the current one." Instead, let the caller decide what the value should be on the first render, or just have it be `null` if the caller is fine with that. [1] zulip#4914 (comment)
As Greg pointed out a few weeks ago [1], `usePrevious` has had a "funny quirk where on the first call, it takes the "previous" value to be the current one." Instead, let the caller decide what the value should be on the first render, or just have it be `null` if the caller is fine with that. [1] zulip#4914 (comment)
As Greg pointed out a few weeks ago [1], `usePrevious` has had a "funny quirk where on the first call, it takes the "previous" value to be the current one." Instead, let the caller decide what the value should be on the first render, or just have it be `null` if the caller is fine with that. [1] zulip#4914 (comment)
Another bundle of these, after #4908, toward #4907.
In general, this turns out to be much easier to do for function components than class components, so most of this series is just doing those conversions. Then there's a commit at the end that adds the annotations.
Temporarily adding
well_formed_exports=true
shows that this series takes us from 205 down to 145 complaints of missing export annotations.