[edit: a proper fix was contributed by someone else in this PR: apache/arrow#14587, my suggested fix below is incomplete].
Minimal reproduction of some Apache Arrow code in Javascript that throws an unexpected error (vector.toArray).
To reproduce the problem:
- Clone the repo
- Run
npm installto install theapache-arrowdependency. - Run
npm run reproto reproduce the problem.
For some reason, the data.reduce operation inside of Apache Arrow's vector.ts is going out of bounds, because the vector's length is 3, but the reduce operation is trying to 2 times 2 things together.
I suppose it could be that the raw IPC bytes might contain a malformed table -- though we've been using this table for a long time, we just never called toArray on it before until now.
Ideally (I would argue), if for (let i = 0; i < vector.length; ++i) vector.get(i); works successfully on a vector, I'd consider the data therein valid, and I'd assume that vector.toArray works as well.
Additionally, the repo contains a Python version of the same code (convert each column in a table into a list, the python equivalent of the Javascript array) which works as expected on the exact same IPC data.
A bit later...
Okay, having debugged the actual code inside Vector.toArray a bit, I'm now inclined to say this might be a bug in the JS Arrow implementation?
Which is to say, the following diff fixes the problem for me -- essentially data.values.length is not always the same as data.length.
I'll go post this on the Arrow Slack channel next to ask if I should submit a PR for this.