Extending by a function output

To work around this issue in Kusto (Azure Data Explorer), you can restructure your approach by using a join instead of extend, as extend only works with scalar values. Since your function returns a table, join can be used to align your datatable with the function output for each row.

Here’s a workaround you can try:

1. Use mv-expand to expand the dynamic array of IDs.
2. Apply the function on each expanded value.
3. Re-aggregate if needed to get the original structure with function results.

Here’s an example of how this can be done:

// Sample function that takes dynamic input and returns a table with one row and two columns.
let function = (ids: dynamic)
{
    datatable(col1: string, col2: string)
    [
        tostring(ids[0]), tostring(ids[1])
    ]
};

// Original table with date and dynamic ids
let original_table = datatable(date: datetime, ids: dynamic)
[
    datetime(2022-01-01), dynamic(["value1", "value2"]),
    datetime(2022-01-02), dynamic(["value3", "value4"])
];

// Expand ids, apply function, and reassemble
original_table
| mv-expand id = ids // Expand to create a row for each id
| extend result = function(dynamic([id])) // Apply the function to each id
| project-away id // Remove intermediate column
| join kind=inner (
    original_table
) on date // Join back with the original table by date

This approach:

• Expands the ids dynamic array so each row has a single value from ids.
• Applies the function on each expanded value.
• Joins back with the original table on a common key (date in this case).

Alternatively, if you have control over the function, you could modify it to return a scalar result based on aggregated data. But if it’s essential to use the function as is, then this join method allows you to align it with your original data.