Skip to content
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

🔥 Stop syncing the value attribute on inputs (behind a feature flag) #13526

Merged
merged 24 commits into from
Sep 12, 2018

Conversation

nhunzaker
Copy link
Contributor

@nhunzaker nhunzaker commented Aug 31, 2018

This PR revives #10150, my original PR to remove value attribute syncing.

TODOs:

  • value set value, defaultValue set defaultValue
  • checked sets checked, defaultChecked sets defaultChecked
  • Remove warning if defaultValue changes (if it exists)
  • Remove warning if defaultChecked changes (if it exists)
  • Can we get rid of wrapper state alltogether?
  • Could we only apply value to text inputs?
  • Could we apply checked to check/radios?
  • Could submit/reset just use standard attribute assignment?
  • SSR mismatch?

This fixture build demonstrates the behavior:

http://react-dom-no-sync-value.surge.sh/


See #11896 and #13525

🔥 🔥 🔥

@nhunzaker nhunzaker changed the title 🔥 Stop syncing the value attribute on inputs 🔥 Stop syncing the value attribute on inputs Aug 31, 2018
@gaearon gaearon self-assigned this Aug 31, 2018
@@ -1605,15 +1605,15 @@ describe('ReactDOMInput', () => {
};
}

it('always sets the attribute when values change on text inputs', function() {
it('retains the initial value attribute when values change on text inputs', function() {
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's always been this way, but I don't like how React stores an initial value:
https:/facebook/react/blob/master/packages/react-dom/src/client/ReactDOMFiberInput.js#L25

I think the value should just be blank when it becomes null, or revert to the defaultValue prop (which could be null).

@pull-bot
Copy link

pull-bot commented Aug 31, 2018

ReactDOM: size: -0.2%, gzip: -0.5%

Details of bundled changes.

Comparing: 2d5c590...5bb2e39

react-dom

File Filesize Diff Gzip Diff Prev Size Current Size Prev Gzip Current Gzip ENV
react-dom.development.js -0.1% -0.6% 645.08 KB 644.23 KB 151.15 KB 150.23 KB UMD_DEV
react-dom.production.min.js -0.2% -0.5% 92.16 KB 91.95 KB 30.08 KB 29.92 KB UMD_PROD
react-dom.development.js -0.1% -0.7% 640.32 KB 639.47 KB 149.75 KB 148.72 KB NODE_DEV
react-dom.production.min.js -0.2% -0.6% 92.1 KB 91.9 KB 29.67 KB 29.51 KB NODE_PROD
ReactDOM-dev.js -0.1% -0.7% 662.65 KB 661.77 KB 152.09 KB 150.97 KB FB_WWW_DEV
ReactDOM-prod.js -0.1% -0.9% 286.89 KB 286.53 KB 53.15 KB 52.68 KB FB_WWW_PROD
react-dom.profiling.min.js -0.2% -0.4% 94.96 KB 94.75 KB 30.33 KB 30.21 KB NODE_PROFILING
ReactDOM-profiling.js -0.1% -0.9% 293.28 KB 292.92 KB 54.51 KB 54.03 KB FB_WWW_PROFILING

schedule

File Filesize Diff Gzip Diff Prev Size Current Size Prev Gzip Current Gzip ENV
schedule.development.js n/a n/a 0 B 19.17 KB 0 B 5.74 KB UMD_DEV
schedule.production.min.js n/a n/a 0 B 3.16 KB 0 B 1.53 KB UMD_PROD

Generated by 🚫 dangerJS

@gaearon
Copy link
Collaborator

gaearon commented Sep 2, 2018

Is there any other logic we can remove or simplify now? I was hoping some of the things we added over time might be unnecessary now. Although I'm aware some of your fixes were for issues that predated #6406.

@gaearon
Copy link
Collaborator

gaearon commented Sep 2, 2018

My original PR also removed synchronization for the checked attribute. However it isn't as problematic, and wasn't mentioned in #13525, so I didn't remove the behavior.

I talked to @sebmarkbage, and the guiding principles we came up with are:

  • React DOM should do what you'd do if you were writing mutations manually.
  • We only support mutations on uncontrolled components.

Our goal is to simplify the system and reduce the edge cases. So if we only set it out of courtesy for e.g. CSS selectors, I think we should stop doing that, and keep it minimal. Just let value set value, defaultValue set defaultValue, and same for checked / defaultChecked.

Regarding form.reset(). We should support it on uncontrolled components. For them, reverting to last supplied defaultValue makes sense. Which would happen anyway. We don't need to fire a change event for that. If you use a reset button, put an onReset on it if you want.

We should not support form.reset() on a controlled component. It is a mutation, and if you want it, just use an uncontrolled component — or change your resetting logic to reset the state instead. So we'll want to warn if you reset a form, and it contains a controlled component inside.

@nhunzaker
Copy link
Contributor Author

Is there any other logic we can remove or simplify now? I was hoping some of the things we added over time might be unnecessary now. Although I'm aware some of your fixes were for issues that predated

I was too. I'll take a second pass through.

For everything else, I've started a checklist in the description of things I need to do.

@gaearon
Copy link
Collaborator

gaearon commented Sep 2, 2018

You might also want to start by adding a feature flag for this. Because we'll want to keep both versions around for a while.

It might be reasonable to fork some files like ReactDOMFiberInput completely later. Let's see which ones have the most feature flag checks after a bit of work.

@nhunzaker
Copy link
Contributor Author

I did some more work today that I believe gets us most of the way there. I shouldn't continue to refine this anymore, next I need to:

  • Break down the changes into distinct feature flags
  • Test a few assumptions about defaultValue since we wrote the Chromium fix for value detachment
  • Follow up on some server side mismatching

I'm really curious about the final item. How should SSR work when value has no influence over the value attribute client-side?

I'm also starting to think that the rules are simple enough now that we can start to think about this more as a spec. I think I'd like to write up some rules, even if it's just for myself.

@nhunzaker
Copy link
Contributor Author

One idea for SSR is to very strictly adhere to "value sets the property, defaultValue sets the attribute". For this to work, we'd need to allow a user to set both properties at once without warning.

I don't necessarily sponsor this idea 😸. This would be a breaking change for just about every server-side app, but it would keep things consistent in both environments.

It also requires a user to be intentional about setting the value attribute for sensitive information, like passwords, which could be a desired side-effect for some users.

Copy link
Collaborator

@gaearon gaearon left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please introduce a new feature flag for this. Let's call it something like disableInputAttributeSyncing. We need to be granular about changes because we might want to try them separately at FB, so I don't want to have just one feature flag.

We want to run relevant tests in both modes. But I think we don't want to run the whole bundle in two modes just yet.

So I propose that for now you can create ReactFire-test.internal.js and copy paste any tests where behavior differs into it, and adjust the expectations. And set the feature flag at the top of this file.

Then we can get it in and start testing at FB. It doesn't have to be 100% complete for this — since it won't be enabled by default. For example SSR stuff can wait.

// manual testing in the fixtures shows that the active element
// is no longer the input, however blur() + a blur event seem to
// be the only way to remove focus in JSDOM
node.blur()
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@gaearon this was the false positive case mentioned earlier. JSDOM doesn't update the active element on blur, only after blur() + a blur event.

I have also walked through the number fixtures to verify that things continue to work as expected manually.

@@ -57,13 +58,14 @@ function isControlled(props) {

export function getHostProps(element: Element, props: Object) {
const node = ((element: any): InputWithWrapperState);
const checked = props.checked;
const checked =
props.checked != null ? props.checked : node._wrapperState.initialChecked;
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You're doing props.checked access twice now. Let's avoid changes like this.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Removed in 14dab53

const currentValue = node.value;
const value = disableInputAttributeSyncing
? getToStringValue(props.value)
: initialValue;
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems like previous we didn't have to call this earlier than next condition. What changed? Let's avoid adding more eager computation.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

getToStringValue is called when assigning initialValue in the wrapper state:

initialValue: getToStringValue(
props.value != null ? props.value : defaultValue,
),

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Something can be done here. Working on it.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was! I did this in addffb8

@@ -63,7 +64,7 @@ export function getHostProps(element: Element, props: Object) {
defaultChecked: undefined,
defaultValue: undefined,
value: undefined,
checked: checked != null ? checked : node._wrapperState.initialChecked,
checked: props.checked != null ? props.checked : node._wrapperState.initialChecked
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I'm being silly. I'll revert this too

return;
}

const initialValue = toString(node._wrapperState.initialValue);
const currentValue = node.value;
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I inlined this value because it is only needed in one place in both feature flag branches.

@nhunzaker
Copy link
Contributor Author

I believe all comments have been addressed and, reading through the code, I feel good about this change.

@gaearon gaearon changed the title 🔥 Stop syncing the value attribute on inputs 🔥 Stop syncing the value attribute on inputs (behind a feature flag) Sep 12, 2018
// 1. The value React property when present
// 2. The defaultValue React property when present
// 3. An empty string
if (initialValue !== node.value) {
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This used to compare toString(node._wrapperState.initialValue) !== node.value.

But now it compares node._wrapperState.initialValue !== node.value.

Is that change safe?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You are correct. This should call toString()

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Needs a regression test?

Copy link
Contributor Author

@nhunzaker nhunzaker Sep 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hard. We only need to call toString for value comparison so that we do not reassign the existing default value and trigger validation. null and undefined are already skipped. This also happens on the post mount step, and not during hydration. I'm not sure how to write a test for it.

We have a manual test fixture for this reason (sorry, I know the formatting isn't ideal):

<Fixture>
<form className="control-box">
<fieldset>
<legend>Empty value prop string</legend>
<input value="" required={true} />
</fieldset>
<fieldset>
<legend>No value prop</legend>
<input required={true} />
</fieldset>
<fieldset>
<legend>Empty defaultValue prop string</legend>
<input required={true} defaultValue="" />
</fieldset>
<fieldset>
<legend>No value prop date input</legend>
<input type="date" required={true} />
</fieldset>
<fieldset>
<legend>Empty value prop date input</legend>
<input type="date" value="" required={true} />
</fieldset>
</form>
</Fixture>
<p className="footnote">
Checking the date type is also important because of a prior fix for
iOS Safari that involved assigning over value/defaultValue
properties of the input to prevent a display bug. This also triggers
input validation.
</p>
<p className="footnote">
The date inputs should be blank in iOS. This is not a bug.
</p>
</TestCase>

export function toString(value: ToStringValue): string {
return '' + (value: any);
export function toString(value: ?ToStringValue): string {
return value == null ? '' : '' + (value: any);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why this change? Did we check every toString callsite to ensure it doesn't break them?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There were two places in postMount where it could stringify null or undefined. Otherwise this can stay the same. Updating now.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My priority here is that the old code path stays as similar as it can be. So that we can merge this without worrying.

Copy link
Contributor Author

@nhunzaker nhunzaker Sep 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Totally. I've done this in e4f68fe


if (ReactFeatureFlags.disableInputAttributeSyncing) {
// TODO: figure out why. This might be a bug.
expect(called).toBe(1);
Copy link
Contributor Author

@nhunzaker nhunzaker Sep 12, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's possible that this happens because value isn't present, so no operation occurs on value, so value tracking is never triggered. A solution could be to set the value from defaultValue or manually call tracking.

Copy link
Collaborator

@gaearon gaearon left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Old code paths look legit.

I'll need to take a closer look at new code paths later but we can get this in and iterate. Thank you!

@gaearon gaearon merged commit a079011 into facebook:master Sep 12, 2018
Simek pushed a commit to Simek/react that referenced this pull request Oct 25, 2018
…acebook#13526)

* 🔥 Stop syncing the value attribute on inputs

* Eliminate some additional checks

* Remove initialValue and initialWrapper from wrapperState flow type

* Update tests with new sync logic, reduce some operations

* Update tests, add some caveats for SSR mismatches

* Revert newline change

* Remove unused type

* Call toString to safely type string values

* Add disableInputAttributeSyncing feature flag

Reverts tests to original state, adds attribute sync feature flag,
then moves all affected tests to ReactFire-test.js.

* Revert position of types in toStringValues

* Invert flag on number input blur

* Add clarification why double blur is necessary

* Update ReactFire number cases to be more explicite about blur

* Move comments to reduce diff size

* Add comments to clarify behavior in each branch

* There is no need to assign a different checked behavior in Fire

* Use checked reference

* Format

* Avoid precomputing stringable values

* Revert getToStringValue comment

* Revert placement of undefined in getToStringValue

* Do not eagerly stringify value

* Unify Fire test cases with normal ones

* Revert toString change. Only assign unsynced values when not nully
jetoneza pushed a commit to jetoneza/react that referenced this pull request Jan 23, 2019
…acebook#13526)

* 🔥 Stop syncing the value attribute on inputs

* Eliminate some additional checks

* Remove initialValue and initialWrapper from wrapperState flow type

* Update tests with new sync logic, reduce some operations

* Update tests, add some caveats for SSR mismatches

* Revert newline change

* Remove unused type

* Call toString to safely type string values

* Add disableInputAttributeSyncing feature flag

Reverts tests to original state, adds attribute sync feature flag,
then moves all affected tests to ReactFire-test.js.

* Revert position of types in toStringValues

* Invert flag on number input blur

* Add clarification why double blur is necessary

* Update ReactFire number cases to be more explicite about blur

* Move comments to reduce diff size

* Add comments to clarify behavior in each branch

* There is no need to assign a different checked behavior in Fire

* Use checked reference

* Format

* Avoid precomputing stringable values

* Revert getToStringValue comment

* Revert placement of undefined in getToStringValue

* Do not eagerly stringify value

* Unify Fire test cases with normal ones

* Revert toString change. Only assign unsynced values when not nully
@burka
Copy link

burka commented Feb 25, 2019

Is there a way to toggle this feature flag without recompiling react?
When is this feature scheduled?

@kripod
Copy link

kripod commented Dec 17, 2019

Is there a reason this hasn't been enabled by default? It seems to be a cause for the bug #17609.

@gaearon
Copy link
Collaborator

gaearon commented Dec 17, 2019

No particular reason other than it's a breaking change so we can't enable it until 17.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

6 participants