Median aggregation overflows for large finite middle values #6654

TanStack Table version

@tanstack/table-core 9.3.1, main at c14e0221b1e656c7bb35e7b3d4cdf188719a3599.

Framework/Library version

Framework-independent aggregation API; reproduced with Node 24.19.0.

Describe the bug and the steps to reproduce it

The even-length median adds the two middle values before dividing by two. That intermediate addition can overflow even when the median is finite and representable.

For example, the median of [Number.MAX_VALUE, Number.MAX_VALUE] is Number.MAX_VALUE, but the current implementation returns Infinity. The negative equivalent returns -Infinity. Unequal large finite pairs have the same issue.

The fix should also preserve tiny/subnormal results: always dividing both operands first would turn the median of [Number.MIN_VALUE, Number.MIN_VALUE] into zero. The existing zero, NaN and infinity behavior should remain unchanged.

Your Minimal, Reproducible Example

Run this with Node 24 from a checkout; this source module only has type imports and needs no framework setup:

import { aggregationFn_median } from './packages/table-core/src/features/row-aggregation/aggregationFns.ts'

const context = values => ({
  rows: values.map(value => ({ value })),
  getValue: row => row.value,
})

console.log(aggregationFn_median.aggregate(context([
  Number.MAX_VALUE,
  Number.MAX_VALUE,
]))) // Infinity; expected Number.MAX_VALUE

Source of the current median calculation.

Four focused Vitest overflow regressions fail on main, while six controls pass (opposite signs, subnormal values, ordinary numbers, signed zero and non-finite inputs).

Do you intend to try to help solve this bug with your own PR?

Yes. I am preparing a small fallback for intermediate overflow and will complete the repository's required checks before opening a PR.