[ARROW-12636] [JS] ESM Tree-Shaking produces broken code

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 4.0.0
• Fix Version/s: 6.0.0, 5.0.1
• Component/s: JavaScript
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/28388
• Assignee: Paul Taylor
• Reporter: Howard Zuo
• Created: 03/May/21 18:45
• Updated: 11/Jan/23 08:27
• Resolved: 30/Jul/21 16:59

Description

When a JS application depends on `apache-arrow` or any of the ESM variants, Arrow's package.json declares `sideEffects: false` which allows bundlers like Webpack or Rollup to remove all code that's evaluated at the top level of all modules (any code that isn't explicitly imported can be pruned).

 

However, there are important top-level side effects that Arrow relies on, e.g. for binding methods on types:

https://github.com/apache/arrow/blob/c0445d69088a6fbd9c026ecac6a99c6cd4df4865/js/src/vector/index.ts#L167-L179

 

So currently, any JS app that uses tree-shaking and depends on `apache-arrow` or the ESM variants will throw if any of those methods are called, e.g. in my case I see `TypeError: e._chunks[t].get is not a function` when calling `table.get`.

 

Indeed switching to the CJS variants like `@apache-arrow/es2015-cjs` fixes the issue, since the Arrow build does not inject `sideEffects: false` into its package.json for non-ESM variants.

 

For reference, the `sideEffects: false` logic was introduced in this PR and issue:

https://github.com/apache/arrow/pull/8418/

https://issues.apache.org/jira/browse/ARROW-10255

 

Issue Links

• links to: GitHub Pull Request #10826