select traces doesn't do partial matches on `meta`

[nicolaskruchten]

Today, the output of

import plotly.graph_objects as go

fig = go.Figure(go.Scatter(meta=dict(a=1,b=2)))

print(len([f for f in fig.select_traces(selector=dict(meta=dict(a=1,b=2)))]))
print(len([f for f in fig.select_traces(selector=dict(meta=dict(a=1)))]))

is 1, 0, meaning that we can't use selector to do partial matches on meta.

It would be really convenient if PX could tag all of its traces with some structured data in meta so that users could target these traces with updates, but for that to be useful we would need to do partial matches.

@jonmmease any major reason not to do this?

[jonmmease]

Does this do what you want?

print(len([f for f in fig.select_traces(selector=dict(meta_a=1))]))

I seem to remember making the design decision that selectors keys needed to be matched in full, and magic underscore notation could be used to target subsets.

Not sure off hand if anything goes wrong if we support partial matches, but it would be a breaking change. We'd also need to decide whether the partial match logic is applied recursively.

[nicolaskruchten]

that doesn't work, no, possibly because meta is an open-ended dict with no schema below it?

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-20-9a7a6c9a1903> in <module>
      4 
      5 print(len([f for f in fig.select_traces(selector=dict(meta=dict(a=1,b=2)))]))
----> 6 print(len([f for f in fig.select_traces(selector=dict(meta_a=1))]))

<ipython-input-20-9a7a6c9a1903> in <listcomp>(.0)
      4 
      5 print(len([f for f in fig.select_traces(selector=dict(meta=dict(a=1,b=2)))]))
----> 6 print(len([f for f in fig.select_traces(selector=dict(meta_a=1))]))

~/plotly/plotly.py/packages/python/plotly/plotly/basedatatypes.py in _perform_select_traces(self, filter_by_subplot, grid_subplot_refs, selector)
    805 
    806             # Filter by selector
--> 807             if not self._selector_matches(trace, selector):
    808                 continue
    809 

~/plotly/plotly.py/packages/python/plotly/plotly/basedatatypes.py in _selector_matches(obj, selector)
    816 
    817         for k in selector:
--> 818             if k not in obj:
    819                 return False
    820 

~/plotly/plotly.py/packages/python/plotly/plotly/basedatatypes.py in __contains__(self, prop)
   3907                     return False
   3908             else:
-> 3909                 if obj is not None and p in obj._valid_props:
   3910                     obj = obj[p]
   3911                 else:

AttributeError: 'dict' object has no attribute '_valid_props'

[nicolaskruchten]

(to be clear: I would be fine with having this work, it would meet my needs just fine :)

[jonmmease]

Ok, yeah. this is special because meta doesn't have a sub-schema.

I think having special case magic underscore logic to handle dicts is a good approach. This would also be useful in the update_* operations for things like:

fig.update_layout(meta_a=8)

It would mean we're asserting that dict keys are strings without underscores. Not sure at the moment if this should be a special case for meta, or if we would want this to work for any property that has a dict value 🤔 But in either case, I think this would be a backward compatible change to make.

[nicolaskruchten]

Or we could have it do partial matches off a dict, as a special case... either way :)

[nicolaskruchten]

On further thought: I want to store keys in meta which are user-provided and likely to have underscores in them...

[nicholas-esterer]

From reading the dialogue above, I gather that select_traces should allow partial matches using meta no matter if it was made using a dict or the underscore notation. Furthermore, meta_a_b=value should produce meta=dict(a=dict(b=value)) but meta=dict(a_b=another_value) should be allowed, too.
Another thing: do you want another keyword in select_traces, say partial_match which allows turning on or off the functionality of matching subsets of trace parameters?

[nicolaskruchten]

I would be OK if partial matches on meta didn't work with magic underscores at all actually, if that's easier to implement.

I don't think I would favour a new partial_match kwarg, no, I think we can just make a special case for meta

[nicholas-esterer]

Ok sounds good.
I don't want to go too far afield here, but what if selector could be a function that returned True if that trace should be selected? The function would be passed the trace. The old dict functionality could be left in, too.

[nicolaskruchten]

that's not a bad idea... I could implement my PX stuff on top of that, yes. Is that easier to build?

[nicholas-esterer]

Maybe, I'm looking into it right now 👍