CPU optimization of model/json

[imix]

Hi

I tried some optimization techniques on dasel and got some improvements.

One of the improvement stems from switching the json library which is maybe not desired. Nevertheless I left the other json library in the pull request to illustrate the potential.

My benchmark was as follows:
cat citylots.json | ./main_orig -i json "features.map(properties.MAPBLKLOT)"

with this json: https://github.com/zemirco/sf-city-lots-json

The version I started with measured as follows (using hyperfine):

Benchmark 1: ./test.sh
Time (mean ± σ): 17.369 s ± 0.032 s [User: 38.968 s, System: 1.464 s]
Range (min … max): 17.324 s … 17.422 s 10 runs

The optimized version had these results:

Benchmark 1: ./test.sh
Time (mean ± σ): 8.440 s ± 0.016 s [User: 14.793 s, System: 1.105 s]
Range (min … max): 8.417 s … 8.472 s 10 runs

Hope this helps or at least gives some insights.

[TomWright]

Hey @imix,

This is great, love to see a contribution like this.
I'll do my best to review and merge over the weekend.

[TomWright]

Hey @imix,

Could I ask you to separate the JSON change into a new PR please? Otherwise it's looking good

[imix]

Hi @TomWright

I reverted the json library in this PR. I will create a second PR with the json library

[TomWright]

I've pulled in new tests for the IsScalar func and the null value doesn't appear to be handled

[TomWright]

I've adjusted the code - it can still likely be improved since we still use isNull, but this still provides a solid improvement:

time cat citylots.json | ./dasel_old.exe -i json 'features.map(properties.MAPBLKLOT)'

21.4s
19.6s
19.2s
18.6s
19.3s

time cat citylots.json | ./dasel_new.exe -i json 'features.map(properties.MAPBLKLOT)'

14.7s
14.8s
15.1s
14.5s
15.2s

I'll merge this here but of course welcome any further improvements.

Thanks for the contribution.

[imix]

Great to see you could implement it! I did not find other big optimization areas so far. Maybe I have a look at Profile-guided optimization if I find the time.