ngrams fails with a bare negative number when the result is too large

[SEPURI-SAI-KRISHNA]

ngrams sizes its result array with int arithmetic, so the size wraps negative for large arrays and the query fails with a bare negative number as its only error message.

Repro

SELECT ngrams(repeat(1, 100000), 50000);

Query failed: -1794917296

That message is the NegativeArraySizeException text leaking through as the query failure; it gives no indication of what went wrong.

Cause

core/trino-main/src/main/java/io/trino/operator/scalar/ArrayNgramsFunction.java

int elementsPerRecord = toIntExact(min(array.getPositionCount(), n));
int totalRecords = array.getPositionCount() - elementsPerRecord + 1;
int[] ids = new int[totalRecords * elementsPerRecord];   // int multiplication

The element count is maximized when elementsPerRecord is about half the array length, giving roughly arrayLength^2 / 4. That exceeds Integer.MAX_VALUE once the array reaches about 92,000 entries. For the repro above:

elementsPerRecord = 50000
totalRecords      = 50001
exact product     = 2500050000
int product       = -1794917296   ->  NegativeArraySizeException

A 100,000-element array is reachable through ordinary SQL, since repeat permits up to MAX_RESULT_ENTRIES = 100_000.

Related existing handling

The sibling functions all guard their result size and fail with a categorized INVALID_FUNCTION_ARGUMENT:

• combinations — checkCondition(combinationCount * (long) combinationLength <= MAX_RESULT_ELEMENTS, ..., "combinations exceed max size")
• repeat — MAX_RESULT_ENTRIES = 100_000
• sequence — MAX_RESULT_ENTRIES = 10_000

ngrams has no equivalent guard.

Environment

Reproduced on master.