Extend dynamicRef keyword

[nezhar]

The description in https://stackoverflow.com/a/69774520/5382179 helped me better understand how the keyword should actually work and the PR extends the missing parts of the current implementation of the keyword.

There might be a better way to implement this, but it would require bigger changes in the way the scopes and the validator functions are handled.

[codecov]

Codecov Report

Merging #886 (12c791e) into main (eed6d8b) will increase coverage by 0.15%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #886      +/-   ##
==========================================
+ Coverage   96.90%   97.05%   +0.15%     
==========================================
  Files          20       20              
  Lines        3259     3290      +31     
  Branches      446      457      +11     
==========================================
+ Hits         3158     3193      +35     
+ Misses         79       77       -2     
+ Partials       22       20       -2     

Impacted Files	Coverage Δ
jsonschema/tests/test_jsonschema_test_suite.py	82.35% <ø> (ø)
jsonschema/_utils.py	97.61% <100.00%> (+0.56%)	⬆️
jsonschema/_validators.py	100.00% <100.00%> (+1.53%)	⬆️
jsonschema/validators.py	97.41% <100.00%> (-0.13%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update eed6d8b...12c791e. Read the comment docs.

[Julian]

@nezhar can you have a look at resolving the conflicts here now that the performance issues with the implementation have been addressed?

[Julian]

Thanks @nezhar. I'm about ready to merge this (sorry for the delay), but to be extra safe I'd like to re-run the benchmarks that @Stranger6667 optimized from #853 -- can you rerun the two they left in that issue there just to be sure we don't regress on them by merging this?

[Stranger6667]

FWIW, some of the work done during ref resolving is only relevant for newer drafts (after Draft 7), so maybe as a follow-up from this PR it would be nice to explore adding some conditional logic to avoid unnecessary work (e.g. the $dynamicAnchor keyword is not a part of Draft 7, so could be skipped during resolving) for older drafts. Not sure how much benefit it would yield though

[nezhar]

@Julian also sorry for the delay 🙂. I made another rebase and run the test from the linked issue. This are my results:

$ python --version; pip freeze | grep json; python test.py
Python 3.8.10
jsonschema==4.3.3
10.12309193611145 sec

$ python --version; pip freeze | grep json; python test.py
Python 3.8.10
jsonschema @ file:///home/hnezbeda/DEV/jsonschema
10.206285238265991 sec

[Julian]

No worries! Will merge I think. Thanks! If you manage to figure out the remaining 2 skipped tests in draft2020 a PR would of course would be very much appreciated. So long as they still fail, my suspicion is that the implementation is still not 100% correct (even in the parts that pass), but it's definitely an improvement. Thanks again!