New rollup version results in different generated output and drops some if/else cases

[digaus]

Rollup Version

2.74.0

Operating System (or Browser)

MacOS, Linux, Windows

Node Version (if applicable)

12.x.x

Link To Reproduction

https://github.com/digaus/rollup-demo

Expected Behaviour

Since 2.74.0 angular lib compilation now produces different and wrong results.

This is a major issue for us and might break other libs and applications because it is not noticed while compiling. It completely drops use cases...

This happens when bundling to FESM2015. In ESM2015 code still looks fine...

Adding "rollup": "2.73.0" to the package.json fixes this for now...

See https://github.com/digaus/rollup-demo for reproduction

Actual Behaviour

Do not drop any code :)

[lukastaegert]

Interestingly, vanilla Rollup does not seem to remove anything here, guess I will need to dig deeper

[costingeana]

Hello, guys,
I observed this behavior in the morning, but I thought I am doing something wrong. Please compare the following code snippets:

• code is removed (it should not): link
• code is not removed: link
• code is not removed: link

I hope they help you to identify the issue(s).

[digaus]

Hello, guys, I observed this behavior in the morning, but I thought I am doing something wrong. Please compare the following code snippets:

• code is removed (it should not): link
• code is not removed: link
• code is not removed: link

I hope they help you to identify the issue(s).

Yes same here. Thought I was too stupid 🤷‍♂️ Refactored code and had the issue from the other issue here. That's how I identified rollup as the cause.

[lukastaegert]

Hello, guys,
I observed this behavior in the morning, but I thought I am doing something wrong. Please compare the following code snippets:
code is removed (it should not): link
code is not removed: link
code is not removed: link
I hope they help you to identify the issue(s).

All of them look correct to me:

• In the first example, f2 does not have a side effect and the return value is not used -> remove
• In the second example, the return is used -> we need to keep it
• In the third example, f2 has a side effect in calling a global function -> keep it

[digaus]

Hello, guys,
I observed this behavior in the morning, but I thought I am doing something wrong. Please compare the following code snippets:
code is removed (it should not): link
code is not removed: link
code is not removed: link
I hope they help you to identify the issue(s).

All of them look correct to me:

• In the first example, f2 does not have a side effect and the return value is not used -> remove
• In the second example, the return is used -> we need to keep it
• In the third example, f2 has a side effect in calling a global function -> keep it

What about my use case? 🙈

[lukastaegert]

I will need more time for the original issue. The repo will help, but I have some family things coming up. Therefore I reverted the main feature from 2.74.0 for now in a 2.74.1 release so that I can take some time to understand the issue properly. Hope 2.74.1 fixes your issue for now? But please keep this one open as I want to roll back the revert soon once I got things figured out.

[costingeana]

Hello again,
Indeed, my first example wasn't good for highlighting the problem I encountered. I apologize.
I attached below 2 more code snippets that are closer to the code where I observed the problem. The code is a bit messy, but I tried to preserve the context.

• parameter is removed: check the call of getParams function
• function is removed (body and call): check the createCollapseButton function

[digaus]

@lukastaegert
2.74.1 seems to work for now, thank you for the fast release 👍

[costingeana]

I can also confirm that 2.74.1 is ok. Thank you.

[lukastaegert]

I created a new PR #4510 that reimplements parameter tree-shaking in a more conservative way. Could you verify that this PR works for you via npm install rollup/rollup#parameter-defaults?

[costingeana]

Hello and sorry for the delay,
I confirm that the PR works for me. Thank you.