map_join is not working correctly when map=geocoder

[alshan]

from lets_plot import *
from lets_plot.geo_data import *

# %%

offices = [
    ["Prague", "CZ", "Headquarters", 100],
    ["Petersburg", "Russia", "R&D Center", 1000],
    ["Moscow", "Russia", "R&D Center", 100],
    ["Novosibirsk", "Russia", "R&D Center", 50],
    ["München", "Germany", "R&D Center", 200],
    ["Amsterdam", "Netherlands", "R&D Center", 100],
    ["Boston", "US", "R&D Center", 10],
    ["Marlton", "US", "Sales", 10],
    ["Foster City", "US", "Sales", 10],
]

dat = dict(
    city=[o[0] for o in offices],
    country=[o[1] for o in offices],
    kind=[o[2] for o in offices],
    size=[o[3] for o in offices],
)

# %%

# Geocoding
city_geocoder = geocode_cities(dat['city']).countries(dat['country'])

# %%

# The map of JetBrains major offices worldwide.
p = (ggplot(dat) +
     # geom_livemap() +
     geom_point(
         # aes(color='kind', shape='kind', size='size'),
         aes(color='kind'),
         size=10,
         map_join = ['city', 'city'],
         map=city_geocoder))
         # map=city_geocoder.get_centroids()))

p.show()

When using map=city_geocoder.get_centroids() it works as expected.

[IKupriyanov-HORIS]

Interesting thing that map=geocoder is actually works as expected.

In your example lets-plot deduces the following:
map_join=[['city', 'city'], ['city', 'country']]
It's because of a single list in map_join that is interpreted as multi-key for data. And as there is no explicit keys for map lets-plot tries to deduce keys by searching known geo-name columns - 'city', 'county', 'state', 'country'.

Sadly there are exactly two matching columns in a map - 'city', 'country'. That's why everything works without errors, but result is incorrect.

Proper way to define map_join for this case is to use pair of lists:
map_join = [['city'], ['city']]

[IKupriyanov-HORIS]

This commit changes map_join parsing policy:

• if map_join is a list of string then it have to have 1 or 2 items, each item is considered as a simple key for data and map.
• if map_join is a list of lists of string then it have to have 1 or 2 nested lists, each nested list is considered as a multi-key for data and map.

Second map_join item can be omitted only if map parameter is a Geocoder or GeoDataFrame, returned by a Geocoder geometries fetching functions.

As result initial code in the issue is now fully working and doesn't require any modifications.