01/18 - Thataway Thursday!

I am so pissed right now and if I ever meet Geoffrey Leonardelli I am going to punch him in the nose for this piece of crap hit I just did. This hit ruined my night. I kept going thinking it has to end, but no, it did not end- I am not even sure how long it took me but had to be close to an hour for 2.00, 2 f'ing bucks..

 

@Kadauchi Been doing some performance testing and I've noticed that most of the popular test cases on jsPerf are useless because they're so simple that the work is simply getting optimized away by optimizing compilers. V8 in Chrome sees a for loop that's accessing variables that are never otherwise used anywhere, for example and says, "hey, cool! we can replace this with a no-op!".



Regular for loops in particular tend to cycle blazingly fast in most jsPerf tests that people put together. I've seen the measure in the millions of ops/sec while the for/of and forEach stagnated in the tens of thosands.

But if you throw in a lot of random data and unpredictable branching that more closely represents real-world cases, the precompiler can't simply walk through and optimize away meaningless for loops. In this one, the for of loop has actually flipped the script and beat the standard for loop by a small margin in every test I've run:

https://jsperf.com/cuylersbmbuster/1

Just a bunch of thinking out loud, and I figured I'd tag you since I see you doing perfomance tests from time to time.

 

The experiment requires that you have headphones or speakers, a working microphone,

 

Title: Technology Choice and Consumer Patterns Survey | Accept Requester: Sam [A3HOXGDFSFPE3Z] Contact TV: [Hrly: $36.02] [Pay: 4.50] [Fast: null] [Comm: null] [Rej: 0] [ToS: 0] [Blk: 0] TO: [Pay: 4.43] [Fast: 5.00] [Comm: 5.00] [Fair: 5.00] [Reviews: 7] [ToS: 0] Reward: $0.50 Duration: 10 minutes Available: 1 Description: This task will ask you your preference between a choice of products as well as choices in consumer patterns. Requirements: Masters Exists; Location EqualTo US

HIT exported from Mturk Suite v2.1.3​

 

Is that why you're still using .includes()?

https://jsperf.com/includes-vs-indexof/4

 

Title: Technology Choice and Consumer Patterns Survey [ $ | $ ] PANDA
Requester: Sam [A3HOXGDFSFPE3Z]
TurkerView: [ $36.02 / hour Generous | Unrated | Unrated ]
Description: This task will ask you your preference between a choice of products as well as choices in consumer patterns.
Duration: 00:10:00
Available: 1
Reward: $0.50
Qualifications:

• Masters Exists
• Location EqualTo US,

[3DASWRA7V4COZ1NXJPRJTPO2M549PR]​

 

Eyewitness Lab [A23G14DF61WLWN] An Eyewitness Memory Study(~ 20 minutes) - $2.50 | PANDA

[33U5EW3SWR804Z6SF0WPWM1EFIV8ZS]​

Good

Unrated

Unrated

$10.80 / hour
00:13:53 / completion time

​

Pros:

Cons:
-Repetitive
-Timed videos so can't really go faster.

To read Eyewitness Lab's full profile check out TurkerView!​

 

mod edit-

 

That's obviously a V8 precompiler no-op optimization. A lot of the native code for the older methods was written specifically to win these synthetic micro-benchmarks back when SunSpider was popular. New code is benchmarked against "real world apps", according to the dev team. Even Octane was deprecated.

Look at what happens when you tweak the context a bit so that the precompiler can't optimize so strongly:

https://jsperf.com/str-indexof-vs-str-includes

Significantly closer performance (to to the point where the difference is insignificant), but declarative readability is maintained more easily with "includes".