Ensure named_objs does not fail on unhashable objects

[philippjfr]

Not all objects are hashable so the named_objs implementation has started causing some errors (e.g. because mpl colormaps aren't hashable anymore). For unhashable objects we try to reverse map the objects by identity.

[codecov-commenter]

Codecov Report

Merging #632 (91399e5) into master (d960230) will decrease coverage by 0.05%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #632      +/-   ##
==========================================
- Coverage   82.91%   82.86%   -0.06%     
==========================================
  Files           5        5              
  Lines        3073     3081       +8     
==========================================
+ Hits         2548     2553       +5     
- Misses        525      528       +3     

Impacted Files	Coverage Δ
numbergen/__init__.py	80.56% <0.00%> (-0.15%)	⬇️

---

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 d960230...91399e5. Read the comment docs.

[jbednar]

y main confusion here is why it even attempts a reverse mapping at all if an explicit namesdict has already been provided.

Even if an explicit namesdict has been provided, it isn't required to be complete. Where that matters is if check_on_set=False for this Selector, in which case someone can do Obj.x=val, for any val, whether that value is in the namesdict or not. The namesdict will then be incomplete, and so named_objs attempts to complete it by providing some representation for every item in the object list. Here you could maybe skip any unhashable items while still attempting to provide a representation for hashable items? But you'd then need the rest of the code to gracefully handle the case of an object not being in the namesdict.

[philippjfr]

But you'd then need the rest of the code to gracefully handle the case of an object not being in the namesdict.

The rest of the code being what in this case? Panel already does and afaik that's the only thing that actually uses the names.

[jbednar]

As revised this seems ok to merge, though really Selector needs to just store proper mappings as outlined in other PRs and issues...