Scroll correction sometimes gets stuck on estimating row sizes #1252
Send to my agentDescribe the bug#
I think this bug is in the same ballpark to #1221:
We have a complex layout that is (for testing purposes) explicitly undersized and causes layout shifts. This sometimes ends up in an infinite loop trying to correct the scroll position and then stays stuck. Internally we've observed this on some of our playwright tests and eventually found a reproduction.
Oddly enough this behavior can be "fixed" by halting the js execution for a frame and then continue, which makes me suspect that the measurement cache or layout measurement is derived from an outdated layout box.
If necessary, I can try to create a specific sandboxed reproduction.
Your minimal, reproducible example#
no reproduction as of yet
Steps to reproduce#
- Have a layout with virtualized lines that sometime wrap lines (and therefore differ from the size in
estimateSize) - scroll up/down with the affected rows appearing/disappearing from the virtualized range
Expected behavior#
Table layout should not jitter.
How often does this bug happen?#
Every time
Screenshots or Videos#
https://github.com/user-attachments/assets/25dbf5de-1f12-442e-8d81-582b1cad8f64
Platform#
- OS: MacOS, Windows
- Browser: Firefox, Chrome
- @tanstack/react-virtual: 3.13.9 ... 3.14.9
tanstack-virtual version#
3.14.9
TypeScript version#
v7.0.0
Additional context#
No response
Terms & Code of Conduct#
- I agree to follow this project's Code of Conduct
- I understand that if my bug cannot be reliable reproduced in a debuggable environment, it will probably not be fixed and this issue may even be closed.
Many scroll-correction fixes are released after 3.14.9. Can you test again with the latest
@tanstack/react-virtual?If the problem is still there, please add a small reproduction, for example on StackBlitz. Without a reproduction, we cannot find the cause.
itisrmk(GitHub profile, opens in a new tab) linked this to #1374
Sign in to join this conversation.
@mio-moto Thanks for the detailed report and the video, could you check:
If it still reproduces after that, a minimal repro would let us dig in. #1221 is unrelated (it's about scrollTo*, and there's no programmatic scrolling here).