Lazy template creation

[titoBouzout]

From time to time, one questions if templates should be created when used, instead of at start-up. Currently, solid creates all the templates at startup, regardless if these are going to be used or not.

In my opinion, templates should be created when used.

Pros:

• pushes time spent in creation to the pages on which the nodes are used, giving a measurable faster startup (in the range of milliseconds).
• from a purist point of view, it doesn't spend time working on things it may not use
• this is specially beneficial for SPA with a lot of components

Cons:

• If pushing the node creation to the pages makes the landing page faster, then that means that switching pages requires creating the nodes, so it will make the navigation slightly slower. The thing is, that usually components are distributed on the pages, so any possible slow down caused by this change will be also distributed and won't be as bad as the slowdown generated by creating all the components at once.

Alternatives:

• make this configurable? Seems not worthy on my opinion.

Please note that this change also includes caching, which means that any template that has the exact same shape will be reused and not regenerated.

Real life testing:

function	templates created	time
current template	153	3ms
improved template	23	0.5ms

[titoBouzout]

I came with the template caching by trying to figure out how to make it lazy. Wasn't intentional, it just happened. I'm not sure if it is worthy, as each call to cloneNode now means it will do map.has, map.get, and then cloneNode, template duplication doesn't seem to happen as frequently. So it's been removed from this pull request.

[ryansolid]

I appreciate the PR and the testing. The plan is to incorporate these improvements assuming they do well in my testing but I've taken it a bit more radically including improvements to compilation and code reduction. Expect to see this in 0.36.x.

Thank you for inspiring this.

[ryansolid]

Good news is this doesn't appear to have any negative impact on simple DOM benchmarks. Like if you show a table with 1000 rows.. and instead you now have to create it one extra time at 1001 it isn't move the dial very much. So that's positive. But lets hope this improves startup and hydration metrics. I think for SSR this is way more interesting because it means that we may never instantiate the template.