Update svelte to 5.0.0-next.174

[trueadm]

There were some bugs in Svelte 165 that affected performance, they should be fixed in this update.

[krausest]

Thanks - I'll post the results tomorrow.

[krausest]

The results are updated. Here's a screenshot (left updated keyed results, right old results): svelte-classic is a bit slower, but it's within the range of noise.

[trueadm]

@krausest That's strange. I get very different results locally – with 174 being significantly faster than before.

[krausest]

I took another look at it just to check if I made some mistake or my update scripts wouldn't work right. So I created two new versions with just the package.json adjusted to .165 and .174, built the two and ran them together with the unchanged keyed/svelte version (i.e. also .174).
I confirmed that the shasum of main.js in keyed/svelte and keyed/svelte-174 was identical and differed from keyed/svelte-165, so I'm quite confident that I didn't mix up versions. Here's the result on my machine (performance looks completely identical for the three versions):

The package-lock of .165 looks like that:

"node_modules/svelte": {
      "version": "5.0.0-next.165",
      "resolved": "https://registry.npmjs.org/svelte/-/svelte-5.0.0-next.165.tgz",
      "integrity": "sha512-r/dr89AAzIsT/tpIJsenhkWjJ7j2Q1Q3tdPu2qgi/8Vc42dtvHPr1XHn9qJhjXIGVE/cS1BxEuwG+Qwp7YUqkw==",

Can you post a screenshot of the results on your machine? What are the specs of your machine?

[trueadm]

I notice that in my runs I get far more variance on all the test cases (select row is +- 1.1). I'm running an M2 Max with 32GB RAM + Chrome 126. So that is probably why the numbers might be off for me. The variance never used to be that bad in earlier versions of Chrome either :/

[krausest]

Just to make sure: Is the testdriver up to date and was it built recently (i.e. after chrome 124)?
(if not => npm run compile in webdriver-ts)

[trueadm]

@krausest I'll do that and re-run the benchmarks for both and also the other frameworks.

[trueadm]

You were right, after updating webdriver-ts, there's almost nothing in it. Interestingly though the overalls core is considerably different from yours and also in comparison to the other frameworks. I assume the scores will normalise out when you do the next browser update?

[krausest]

That's good!
This is something I have no good explanation. Incremental runs (just the updated frameworks + vanillajs) tend to perform worse than full runs. I eliminated minor browser updates by switching to a fixed chrome for testing version, I'm stopping onedrive and I'm quitting all apps whenever I'm running the benchmark.
For a full run I'm also rebooting before running and I guess there are less background processes (like karabiner elements, betterdisplay, etc. won't be started), but I'm not sure what is guilty here. Seems to have become worse since a few months.
The last full run was closer to your results. So I indeed hope that results look better in a full run.

P.S. I'm surprised the absolute numbers from the full run for my M1 are so close to your M2.

[krausest]

If you want to read more about the chrome 124 issue, this is a good start: #1661 (comment)

[trueadm]

@krausest M1 and M2s are sensitive to temperature. I have a window near my Mac which helps keep it cool. If I close my window and its warm outside, the results jump all over the place pretty quickly (you can use a core speed tracker).

[krausest]

I'll have to keep an eye on that. When I had a MacBook Air M1 I looked at powermetrics: #885 (comment) but didn't find signs of throttling (at least that way).
Maybe I'll take a look at Mx Power Gadget. What do you use to find out about throttling?