Should we add a variant built by non-webpack module bundler?

[tetsuharuohzeki]

Webpack is still one of the popular module bundler in JavaScript ecosystem. But some other tools also emerges.

For example, there is a rollup. Rollup is not only used as standalone tool, it' also used as vite's backend.

rollup emit a code that all modules are flatten by analyzing a module graph deeply. This is contrast to webpack wraps each modules by function as a emulate a module scope.

I'm not sure about that whether JSVM has already optimized about that this kind of a code emitted by rollup. However, in today's web application size, their flatten code including many variables might have some impacts related to web browser's page initialize performance.

I don't think we should drop webpack variant but I also think it might be better to add more module linker variants to get a chance to improve more workloads.

[flashdesignory]

This is a great discussion point.
The initial goal of the todoMVC workload updates is to introduce some consistency between the workloads, to be able to determine impact of frameworks / libraries on performance.

Currently the todoMVC workloads use a variety of module bundlers and it makes it harder to compare.
I do agree that it would be useful to determine if bundlers have impact, for example a react workload built with webpack vs a react workload built with rollup, parcel or any other bundler.

This could potentially also be an iteration, once the initial updates have landed?

cc: @kara @housseindjirdeh

[rniwa]

Note that we specifically don't want to able to compare different frameworks' performances. That is specifically a non-goal.

[tetsuharuohzeki]

Note that we specifically don't want to able to compare different frameworks' performances. That is specifically a non-goal.

Yes. I opened this issue to discuss about to complement more variants of typical web application workload in today.

[bgrins]

FWIW the new rich editor workloads from #81 are using Vite/Rollup. Also when prototyping some changes to todo workloads I found Parcel helpful (though this isn't part of the actual work that's going to land).

I would be curious to learn if the compiled code from these different tools actually have a strong Speedometer-measurable effect. Given that we start measuring after the page and all scripts are loaded my assumption is no - and that we shouldn't create permutations of workloads just to isolate that component - but it wouldn't hurt to check.

[julienw]

The new svelte implementation also uses rollup.

[julienw]

The proposed charting workload uses vite/rollup as well.

[rniwa]

Sounds like this is taken care of now. Is there anything else we should discuss before closing this issue?

[tetsuharuohzeki]

I don't have any additional topic. Thank you for your discussion.

[rniwa]

Closing the issue as complete since we have workloads that use vite/rollup now.