-
-
Notifications
You must be signed in to change notification settings - Fork 3k
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
fix: performance improvements on input widgets #6565
base: main
Are you sure you want to change the base?
Conversation
aa04fad
to
aee2cb7
Compare
✅ Deploy Preview for decap-www canceled.
|
…content; improved perf ftw
a226152
to
44f3d99
Compare
Hey @geotrev this looks very cool and we'd love to merge the PR. Would you mind resolving the conflict that the package rename caused? |
✅ Deploy Preview for cms-demo ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
a528208
to
798921d
Compare
Thanks @demshy, looks like some change sare being pushed by @martinjagodic is also pushing some changes. I'm not sure where the code is at, so I will make time to review the head of the branch shortly so I know what's going on with it. :) |
✅ Deploy Preview for decap-www canceled.
|
@demshy I've updated datetime and rechecked the other input-based components, and it looks good to go from my end. let me know if there are any other concerns and I can address them as needed. Also removed Markdown widget from scope. Also, something I'm now realizing after making this PR is that, basically all these components fully ignore This goes back to the original issue in this CMS which is that all content was centrally managed by redux, which is the cause of the intense delay on input in the first place. As long as the content can be "trusted" to match the UI, then I still stand by this PR, but it's just another thing to consider on the overall architecture of the CMS and solve for in the future. Though y'all may already be on top of it, and this PR may be the lesser of two evils, so forgive me if I'm spouting redundancy. :D |
Late night inspiration for decap devs: I actually really like the idea of experimenting with global state separating from ui state overall, as long as the checks are robust and frequent enough to ensure UI and final CMS state are in sync. the key of course is probably engineering a solution that can receive the CMS stored value & reflect it in relevant text input fields, periodically. Perhaps through a custom hook that can mix the controlled |
Hey there, sorry for taking so long, but a combination of circumstances kept me off the project for longer than I anticipated. We are actually contemplating a complete rework of state management and will consider your insights when the discussion resumes (or even better we open up a thread on discord). For now, I will merge the PR once the tests go through. Thank you for this! |
Seems like a few more clock advances will be needed in e2e tests now with the debounce. Calling |
Darn. I can try and fix those locally when I have more time, if you haven't gotten to it. |
I have pushed an update that fixes some of the tests, You can reproduce the bug like this:
|
@demshy that's great! Anything else I can help with? Looks like CI is still failing, happy to jump on a bandwagon if needed. |
Oh right, I noticed now that I crossed out the text in a way that made it ambiguous 😅 The bug with field_validations_spec that I described is still there, because I haven't been able to get around to it yet. |
Apologies, I also haven't been able to get to this. Not sure when I will. Will update if that changes! |
@demshy So I ended up having time immediately as I posted my previous comment. I ended up reverting the list/object widget changes since I don't think they have a significant performance impact compared to the input widgets. I see the cypress tests for field_validations pass locally, so let's see if they pass in CI! |
Any update on this? |
fixes #3415.
Summary
Update 5/8/2024: List/Object widget changes were removed from the scope of this PR as they aren't relevant to the input performance changes needed, and have an overall minimal impact on performance anyway. If it becomes a huge concern it can be fixed in a follow on PR.
Performance of list and object widgets degrade quickly given large sub-widget composition. See above issue with reproductions.
After pondering and exploring this on and off for a few weeks, I've concluded the issue is largely to do with how the app is providing the onChange handler for input widgets (among others). The onChange handler then updates on every key stroke, which is bad as Redux runs synchronously, and for such a large app, that will cause delays/stuttering when typing.
A separate, but related, issue is that collapsed list/object widgets are still rendering despite being visibly hidden. It's a minor perf improvement but any reduction in reconciliation is a win, IMHO, so I went ahead and hid those component trees where appropriate.Proposed fixes
Refactor each input widget to rely on self-contained React state for their value, then call the provided
props.onChange
on a 300ms trailing debounce to save changes back to the Redux store. I used 300ms to account for relatively slow typing, but am open to suggestions on other increments.Also updated existing debounce durations on the Markdown widget for consistency.Affected widgets:
Markdown(removed from scope)Prevent collapsed List and Object widgets from rendering when not in opened state. This is a much smaller improvement to clean up the DOM and reduce React reconciliation time (noticable delay is noticed on widget expansion without this fix).Test plan
Fix 1: The main way I'm testing this is by loading up the kitchen sink collection and testing responsiveness of form fields with large collections of inputs and visible list items. Then refreshing and making sure unsaved changes can be re-applied from localStorage (I believe that's what the CMS uses to retain those changes). Added
jest.runAllTimers
to relevant unit tests to fulfill existing test cases.Fix 2: Manual testing for speed improvements. Tests that used to assert against now-removed DOM nodes in collapsed state are removed and replaced by tests checking a null render.Otherwise, relevant unit tests for each widget have been updated.
Unit tests:
Checklist
yarn format
.yarn test
.🐶 🐱 🐰 🦔 🐀 🐁 🐜