Error when query contains a UNION value

[jraymakers]

What happens?

When the following query is run in DuckDB WASM:

from (values (union_value(str := 'abc')));

an error like the following is returned:

Cannot destructure property 'length' of '(intermediate value)(intermediate value)(intermediate value)' as it is undefined.
    at VectorLoader.readData (vectorloader.ts:125:56)
    at VectorLoader.visitUtf8 (vectorloader.ts:70:151)
    at VectorLoader.visit (visitor.ts:26:48)
    at VectorLoader.visit (vectorloader.ts:54:22)
    at visitor.ts:23:44
    at Array.map (<anonymous>)
    at VectorLoader.visitMany (visitor.ts:23:22)
    at VectorLoader.visitSparseUnion (vectorloader.ts:103:150)
    at VectorLoader.visit (visitor.ts:26:48)
    at VectorLoader.visit (vectorloader.ts:54:22)

This happens when the query is passed directly to the DuckDB WASM JS API (i.e. query on AsyncDuckDBConnection).

In shell.duckdb.org, running the query above does not return any results, either an error or the correct output.

In the native CLI, the query above returns:

D from (values (union_value(str := 'abc')));
┌────────────────────┐
│        col0        │
│ union(str varchar) │
├────────────────────┤
│ abc                │
└────────────────────┘

To Reproduce

In the esbuild-browser example, replace the query with from (values (union_value(str := 'abc')));.

Browser/Environment:

Chrome 118

Device:

M2 MacBook

DuckDB-Wasm Version:

1.28.0

DuckDB-Wasm Deployment:

local (esbuild-browser example)

Full Name:

Jeff Raymakers

Affiliation:

MotherDuck

[jraymakers]

This does not appear to depend on the type of the value in the UNION. For example, using an integer results in the same error with a slightly different stack trace:

from (values (union_value(num := 123)));

Cannot destructure property 'length' of '(intermediate value)(intermediate value)(intermediate value)' as it is undefined.
    at VectorLoader.readData (vectorloader.ts:125:56)
    at VectorLoader.visitInt (vectorloader.ts:64:113)
    at VectorLoader.visit (visitor.ts:26:48)
    at VectorLoader.visit (vectorloader.ts:54:22)
    at visitor.ts:23:44
    at Array.map (<anonymous>)
    at VectorLoader.visitMany (visitor.ts:23:22)
    at VectorLoader.visitSparseUnion (vectorloader.ts:103:150)
    at VectorLoader.visit (visitor.ts:26:48)
    at VectorLoader.visit (vectorloader.ts:54:22)

[jraymakers]

Also note that this currently causes all_types.test.ts to fail:

  Test All Types Async
        OK: types with custom config (13ms)
        OK: partially supported types (11ms)
    FAILED: fully supported types (13ms)
        TypeError: Cannot destructure property 'length' of '(intermediate value)(intermediate value)(intermediate value)' as it is undefined.
            at VectorLoader.readData (packages/apache-arrow/visitor/visitor/vectorloader.ts:125:56 <- packages/duckdb-wasm/dist/tests-browser.js:14648:23)
            at VectorLoader.visitInt (packages/apache-arrow/visitor/visitor/vectorloader.ts:64:113 <- packages/duckdb-wasm/dist/tests-browser.js:14580:111)
            at VectorLoader.visit (packages/apache-arrow/visitor.ts:26:48 <- packages/duckdb-wasm/dist/tests-browser.js:8859:46)
            at VectorLoader.visit (packages/apache-arrow/visitor/visitor/vectorloader.ts:54:22 <- packages/duckdb-wasm/dist/tests-browser.js:14571:20)

[jraymakers]

I'm pretty sure this is caused by this bug in apache-arrow, which was fixed in version 14. So upgrading apache-arrow should fix this.

[jraymakers]

I've confirmed that upgrading to apache-arrow v14.0.1 (the latest at the time of this writing) fixes this issue.