Skip to content
This repository has been archived by the owner on Oct 8, 2021. It is now read-only.

function next(eit::EdgeIter, state) now does not modify state #423

Merged
merged 4 commits into from
Aug 29, 2016

Conversation

IssamT
Copy link
Contributor

@IssamT IssamT commented Aug 29, 2016

Fixing #419

It seems that there are some exceptions in julia Base where next modified its arguments, but I don't think it is needed here. Maybe if you have some benchmarks I can run them and see the impact of such modification.

@codecov-io
Copy link

codecov-io commented Aug 29, 2016

Current coverage is 98.14% (diff: 100%)

Merging #423 into master will increase coverage by <.01%

Powered by Codecov. Last update da2a26f...48817c4

@sbromberger
Copy link
Owner

Can you run a quick benchmark before and after?

g = Graph(10_000,1_000_000)
@time z = collect(edges(g));

ought to test things.

@IssamT
Copy link
Contributor Author

IssamT commented Aug 29, 2016

There are more allocations as expected. Maybe we should keep it as it is without my modifications:

after patch:

0.087874 seconds (1.02 M allocations: 23.315 MB, 19.94% gc time)
0.049320 seconds (1.00 M allocations: 22.888 MB, 21.18% gc time)
0.085568 seconds (1.00 M allocations: 22.888 MB, 32.56% gc time)
0.124874 seconds (1.00 M allocations: 22.888 MB, 67.30% gc time)
0.069871 seconds (1.00 M allocations: 22.888 MB, 27.71% gc time)
0.141565 seconds (1.00 M allocations: 22.888 MB, 67.78% gc time)
0.096521 seconds (1.00 M allocations: 22.888 MB, 28.45% gc time)
0.108220 seconds (1.00 M allocations: 22.888 MB, 66.77% gc time)
0.085522 seconds (1.00 M allocations: 22.888 MB, 26.22% gc time)
0.147368 seconds (1.00 M allocations: 22.888 MB, 71.82% gc time)
0.086941 seconds (1.00 M allocations: 22.888 MB, 24.80% gc time)
0.114960 seconds (1.00 M allocations: 22.888 MB, 70.63% gc time)
0.100419 seconds (1.00 M allocations: 22.888 MB, 28.55% gc time)
0.117108 seconds (1.00 M allocations: 22.888 MB, 67.41% gc time)
0.097648 seconds (1.00 M allocations: 22.888 MB, 27.85% gc time)
0.131176 seconds (1.00 M allocations: 22.888 MB, 69.38% gc time)
0.079808 seconds (1.00 M allocations: 22.888 MB, 27.42% gc time)
0.132156 seconds (1.00 M allocations: 22.888 MB, 66.04% gc time)
0.095013 seconds (1.00 M allocations: 22.888 MB, 26.51% gc time)

before patch

0.079939 seconds (16.54 k allocations: 8.038 MB, 4.63% gc time)
0.059864 seconds (8 allocations: 7.630 MB, 14.42% gc time)
0.071052 seconds (8 allocations: 7.630 MB, 11.16% gc time)
0.057030 seconds (8 allocations: 7.630 MB, 8.51% gc time)
0.045685 seconds (8 allocations: 7.630 MB, 6.36% gc time)
0.056548 seconds (8 allocations: 7.630 MB, 9.12% gc time)
0.042988 seconds (8 allocations: 7.630 MB, 7.70% gc time)
0.057634 seconds (8 allocations: 7.630 MB, 6.93% gc time)
0.055834 seconds (8 allocations: 7.630 MB, 7.63% gc time)
0.049039 seconds (8 allocations: 7.630 MB, 5.99% gc time)
0.048108 seconds (8 allocations: 7.630 MB, 6.92% gc time)
0.050112 seconds (8 allocations: 7.630 MB, 6.60% gc time)
0.052723 seconds (8 allocations: 7.630 MB, 7.71% gc time)
0.057187 seconds (8 allocations: 7.630 MB, 6.31% gc time)
0.066122 seconds (8 allocations: 7.630 MB, 8.54% gc time)
0.060069 seconds (8 allocations: 7.630 MB, 6.25% gc time)
0.056990 seconds (8 allocations: 7.630 MB, 6.36% gc time)
0.041709 seconds (8 allocations: 7.630 MB, 9.59% gc time)
0.051158 seconds (8 allocations: 7.630 MB, 7.83% gc time)
0.063857 seconds (8 allocations: 7.630 MB, 7.48% gc time)
0.059108 seconds (8 allocations: 7.630 MB, 7.01% gc time)
0.054599 seconds (8 allocations: 7.630 MB, 8.30% gc time)
0.048916 seconds (8 allocations: 7.630 MB, 6.82% gc time)
0.057066 seconds (8 allocations: 7.630 MB, 7.52% gc time)
0.051359 seconds (8 allocations: 7.630 MB, 5.99% gc time)
0.054086 seconds (8 allocations: 7.630 MB, 6.77% gc time)
0.055866 seconds (8 allocations: 7.630 MB, 8.53% gc time)
0.070306 seconds (8 allocations: 7.630 MB, 6.70% gc time)
0.051493 seconds (8 allocations: 7.630 MB, 7.85% gc time)
0.061646 seconds (8 allocations: 7.630 MB, 7.22% gc time)

@sbromberger
Copy link
Owner

Oh, yeah, look at those allocations. Not sure this is a great idea right now, though I understand why we'd want to consider it. Thanks for pulling it together though.

@IssamT
Copy link
Contributor Author

IssamT commented Aug 29, 2016

I don t see surprising info in the amount of allocations. Returning a separate state when iterating 1M times requires 1M allocations. the time doubles in average which is less than what I expected from the allocations. But yes if the package aims to reach the performance edge, this is definitely not a good idea, so feel free to close both the issue and the PR.

@IssamT
Copy link
Contributor Author

IssamT commented Aug 29, 2016

using immutable

0.085918 seconds (15.47 k allocations: 8.019 MB, 4.61% gc time)
0.030574 seconds (7 allocations: 7.630 MB, 21.09% gc time)
0.025852 seconds (7 allocations: 7.630 MB, 14.91% gc time)
0.037600 seconds (7 allocations: 7.630 MB, 7.08% gc time)
0.046564 seconds (7 allocations: 7.630 MB, 6.63% gc time)
0.040199 seconds (7 allocations: 7.630 MB, 5.60% gc time)
0.036929 seconds (7 allocations: 7.630 MB, 4.33% gc time)
0.030815 seconds (7 allocations: 7.630 MB, 5.98% gc time)
0.025692 seconds (7 allocations: 7.630 MB, 8.37% gc time)
0.035490 seconds (7 allocations: 7.630 MB, 4.58% gc time)
0.030208 seconds (7 allocations: 7.630 MB, 8.50% gc time)

EDIT:

The results are even better than before... I m surprised.

@sbromberger
Copy link
Owner

Now those speeds are compelling :)

@jpfairbanks
Copy link
Contributor

These numbers look close enough to existing performance.

@sbromberger
Copy link
Owner

sbromberger commented Aug 29, 2016

Where's the code, though?

Edit: ah, there we go.

@jpfairbanks
Copy link
Contributor

Haha #WheresTheresTheCode

s = state.s
di = state.di + 1
di = state.di
first || (di += 1)
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please rewrite this without short-circuiting (per our guidelines):

Style point: prefer the short circuiting conditional over if/else when convenient, and where state is not explicitly being mutated (e.g., condition && error("message") is good; condition && i += 1 is not).

@sbromberger sbromberger merged commit 2e54807 into sbromberger:master Aug 29, 2016
@sbromberger
Copy link
Owner

@IssamT - thanks for your help with this!

@IssamT
Copy link
Contributor Author

IssamT commented Aug 29, 2016

My thanks to you @sbromberger for helping make the fix better that my first suggestion

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants