Update Svelte version

[trueadm]

As per the title, comes with some API changes too.

[krausest]

Here's the incremental run:

I ran on AC, but vanillajs was still slower than in the last full run, so with that as a reference I'd say no change for svelte. Maybe I'll try later and quit some background processes and results get closer to a full run. see #1721

[krausest]

Here's a screenshot after a reboot on AC, most background processes stopped:

vanillajs got faster, but it didn't help svelte.

[trueadm]

Thanks for that. Separate note:

Is there something we can do with the really high variance on "select row"? I've been noticing it locally too, each run will result in a value that is almost +/- 2, which skews the geometric mean each time. Maybe that particular benchmark could run more times and exclude more outliers to reduce the variance?

[krausest]

select row is indeed painful.
I'm already using two measures:

1. select rows takes 25 samples instead of 15 as for the other benchmarks
2. the weighted geometric mean reduces the influence of select row (see https://github.com/krausest/js-framework-benchmark/wiki/Computation-of-the-weighted-geometric-mean)

Here's an example for svelte and select rows what the average score would be - and it's pretty close thanks to the weight:

Example duration for select rows	factor value / fastest	total weighted score
4.9	1.88	1.162694559
3.9	1.5	1.150468554
5.9	2.27	1.17273775

Another check: A duration between 4.3 msecs and 5.1 msecs yields a total result of 1.16.

With a weight of 1 for each benchmark the total score would be 1.18 and 1.23 for 3.9 and 5.9. So I really think the weighted average makes sense.

Attached you find an excel to play with the results.

Select rows is just too fast to find a winner, but it can identify slow frameworks.

results.xlsx

[trueadm]

That definitely makes a lot of sense. Does making it 6x slowdown have any improvement on the metric if it's already so quick?

[krausest]

I'll take a look at it when I'm back from vacation.