check for sideEffects field set to false

[KarishmaGhiya]

cc : @ramya-rao-a I am not sure if the sideEffects field in core-asynciterator and core-paging were set to true intentionally. Can you ask the appropriate people to review this PR?

[sophiajt]

We shouldn't change paging and the polyfill module, as these do have side effects. We might be able to limit it to one. We can revisit after the release.

[KarishmaGhiya]

@jonathandturner Reverted the changes for core-paging and core-asynciterator

[ramya0820]

This seems to be a setting mostly used by WebPack. Since rollup doesn't use it and has it's own mechanism, doesn't seem like we should be adding this. More context -rollup/rollup#2593
Deferring to @bterlson for any other insights

[chradek]

Our customers may be using the package with webpack, which is why it can be useful to have this field.

[bterlson]

Yes this is for customer bundling more than our own. Webpack is far and away the most popular tool for this.

[chradek]

As far as I know, core-amqp doesn't contain any side-effects.

Side-note: Do we have a way to test if our package is actually side-effect free? I wonder if creating a bundle of our samples and the package using webpack and running the samples would catch issues from a bad configuration (it's what our customers would likely be doing)

[southpolesteve]

Approved for Cosmos. We are side effect free and I kept meaning to add this. Our webpack users will benefit. @chradek I like the idea of having some tool that can help verify this. I am not aware of anything, but maybe @bterlson has already given it some thought.

[bterlson]

Confirming the polyfill libs should have sideEffects: true.

amqp-common seems right to me.

I don't know how to scan for side effects. Theoretically speaking it requires an effects tracking type system which would be a fun project but outside the scope of this project 😄 Practically speaking one could probably hack something into Node to track certain classes of side effects (e.g. any writes to types in global scope during module load) but this would likely never catch all side effects and be a lot of work besides.

Testing that the bundles we produce can be composed into working applications using a variety of bundlers seems like the best way to validate this.

The review comment has been addressed. Dismissing to unblock the PR