♻️ avoid deepcopy of dict in validate_coerce

[jvdd]

This PR avoids the deepcopy in the _plotly_utils.basevalidators.BaseDataValidator its validate_coerce method.

The reason I updated this code; I profiled plotly.py its graph construction time when adding large traces and +/- 70% of the time was spent in this validate_coerce method - of which +/- 33% of the time was spent at the deepcopy (see below).

Line #      Hits         Time  Per Hit   % Time  Line Contents
==============================================================
  2626                                               def validate_coerce(self, v, skip_invalid=False, _validate=True):
  2627         2         26.0     13.0      0.0          from plotly.basedatatypes import BaseTraceType
  2628                                           
  2629                                                   # Import Histogram2dcontour, this is the deprecated name of the
  2630                                                   # Histogram2dContour trace.
  2631         2        321.0    160.5      0.1          from plotly.graph_objs import Histogram2dcontour
  2632                                           
  2633         2          5.0      2.5      0.0          if v is None:
  2634         1          2.0      2.0      0.0              v = []
  2635                                                   else:
  2636         1          4.0      4.0      0.0              if not isinstance(v, (list, tuple)):
  2637                                                           v = [v]
  2638                                           
  2639         1          3.0      3.0      0.0              res = []
  2640         1          1.0      1.0      0.0              invalid_els = []
  2641         2          5.0      2.5      0.0              for v_el in v:
  2642                                           
  2643         1          3.0      3.0      0.0                  if isinstance(v_el, BaseTraceType):
  2644                                                               # Clone input traces
  2645         1     138436.0 138436.0     44.4                      v_el = v_el.to_plotly_json()
  2646                                           
  2647         1          5.0      5.0      0.0                  if isinstance(v_el, dict):
  2648         1      99891.0  99891.0     32.0                      v_copy = deepcopy(v_el)

I believe it is unnecessary overhead to first create a dict (with .to_plotly_json) and then make a deepcopy of that dict to possibly pop one key (even when v_el is a dict and not a BaseTraceType, this still seems like unnecessary overhead to me).

Would appreciate any remarks!

Perhaps it might also be interesting to try avoid the .to_plotly_json call when it is a BaseTraceTtype - as this also takes a considerable amount of time (for large traces)?
-> @jonasvdd took a look at this and replaced the .to_plotly_json (which calls deepcopy on the self._props) with just self._props)

---

What I profiled:

x = np.arange(5_000_000)

%lprun -f plotly.graph_objects.Figure.add_trace plotly.graph_objects.Figure().add_trace(go.Scatter(x=x, y=x))

Before & after this PR:

[jvdd]

I believe this check never hits as in in line 2647 is checked whether v_el is a dict.

[nicolaskruchten]

Thanks for this PR! Sorry I haven't had a chance to review it in detail yet, but the performance gains seem promising!

[jonasvdd]

Hi @nicolaskruchten,

No worries, we both know how busy times can get!

Cheers,
Jonas & Jeroen

[alexcjohnson]

💃 Thanks very much @jvdd, and apologies for the delay!