JSON.parse performance about argo

[tonyd256]

This is cool, thanks for doing it! This looks like the master branch right?

[piercifani]

Yep, Argo v0.4.2 from CocoaPods. I just saw that you pushed 0.4.3, do you want me to re-run the tests again?

[tonyd256]

We did some testing. Using a 85KB JSON file we made we are seeing ~160ms to run JSON.parse. Messing around a bit we discovered some optimizations that can bring this down to ~130ms. We will make this changes for sure.

That being said, we are running our tests under the Release scheme which is configured for a Optimization level of Fastest [-O]. When I changed that to None [-Onone] we saw a large increase in execution time. ~360ms with the code as is on master and ~177ms with the optimizations we found. If you are running this in Instruments do you think it's being build to the debug scheme and therfore has an Optimization level of None [-Onone]? I'm curious to see if your tests are running with Fastest [-O] and if not, what times are you seeing if you make that change.

[tonyd256]

Also, to answer your question, 0.4.3 should not make a difference in execution time of that function.

[piercifani]

The results I mentioned are with the DEBUG scheme. However, I also tried running it on my phone under the Release scheme and all compiler optimizations turned in, and it did bring it down from 620ms to 450ms or around there. It's still way higher than what it should be, but I can't seem to figure out what the bottleneck is...

[tonyd256]

That's good to know that the Release scheme is helping a bit. We have an optimization that should cut that down even more and we'll push it next week. The big problem here is the lack of native support around combining dictionaries. Because of this, the compiler isn't optimizing much for us. I think future improvements to Swift will help with this. Also, you might want to look at cutting down the size of the JSON response as large responses like this can take a noticeable amount of time to download in slow network areas.

[piercifani]

Yes, trimming down the JSON size is one of our main objectives right now. But my goal opening this issue was pointing out the huge performance difference between Argo and competing solutions. I'd gladly pay a 2x performance hit if I get all the huge benefits I get by using Argo, but 10x it's hard to justify to my teammates.

Let's hope that next week's release with your improvements will help performance and further Swift enhancements will give us better tools to build faster code.

All the best!

[tonyd256]

@piercifani ... #88 addresses this issue. If you can try that out and let me know how it goes that would be great.

[piercifani]

I couldn't install the td-faster branch using CocoaPods since it looks like there are some dependencies that are not specified in the Podspec yet.

However, I was able to copy-paste the new JSON.parse implementation into my project and try it out.

These are the results on the Simulator:

Build Config	Branch	Time to parse
Debug	master	0.195s
Debug	td-faster	0.192s
Release	master	0.183s
Release	td-faster	0.092s

So there is a big improvement on the Release build, awesome!

[piercifani]

However, when testing on a device, I get a nasty EXC_BAD_ACCESS while parsing if I build with Release, weirdest crash ever....

[tonyd256]

A lot of what's on master is for Swift 1.2 beta3 ... if you're using it in Swift 1.1 or even an earlier beta then crashes are very possible. Anyways, that's great that the speed improved. I'll close this issue when the PR is merged and we'll continue to seek better performance with newer versions of Swift.

[piercifani]

No, that can't be it, I'm building with latest Xcode beta, with latest Swift version:

pcifani$ swift -v
Apple Swift version 1.2 (swiftlang-602.0.44.6 clang-602.0.45)
Target: x86_64-apple-darwin14.1.0

However, awesome, please close as soon as you merge the new code, as it looks like these issues are resolved.

[tonyd256]

The PR code is now on master.