You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Sorting puts NaN and invalid dates last in either direction, like missing values, instead of leaving the order to the engine's sort.
A column that mixes numbers, dates and text sorts by kind (numbers, dates, text) so the order is consistent, and bigints sort by value.
A page size of NaN reads as 10 and Infinity puts every row on one page; setPageIndex(NaN) and setPageSize(NaN) no longer write NaN into the state.
getFacetedMinMaxValues ignores NaN values.
Per-row selection and expansion cells are dropped for rows that left the data (a table whose row ids keep changing no longer grows for its whole life); a row still in the data, or a group row, keeps its cell and its state.
Building columns and grouping no longer recurse over the nesting depth, so column groups and grouped columns can be nested thousands deep.
A selection or expansion flag that an effect or memo is reading is no longer dropped when its row leaves the data: an observer of a row that leaves and comes back keeps hearing of its changes. Unwatched flags are swept as before.
setPageSize after an Infinity page size with a page index above 0 (written through setPagination) goes to the first page instead of storing an Infinity or NaN page index.