Choropleth selectors: one region at a time, in the order the trace declared
them.
Plotly draws a path per calc entry — including the entries it resolved to no
region at all, which stay in the DOM with no shape — so a region is
addressed by its own position rather than by counting the ones before it.
That keeps the pairing right for a map whose data has holes, which is the
ordinary case: a value the source had nothing for is still a row.
The trace's group is counted within its own geo subplot, since plotly hangs
no uid class on these groups either.
The subplot is named by its whole class attribute, not by a class token.
Plotly writes a geo subplot's attribute as "geo " + id, so the first one
reads class="geo geo" — a doubled string whose classList is the single
token ["geo"], which is a token every other geo subplot carries as well
(class="geo geo2" → ["geo", "geo2"]). Measured in Chromium on a
two-map figure: .geolayer > g.geo and .geolayer > g.geo.geo are the same
selector and both matched 2 of 2 subplots, so every region selector scoped
to geo resolved to one path in every map — region 1 came back as USA
and FRA at once. g[class='geo geo'] matched exactly 1, and attribute
matching is whole-string, so geo10 collides with neither geo nor geo2.
What that cost was the first map's whole highlight.
ChoroplethTrace.mapToSvgElements compares the resolved elements against
the declared regions and returns null on any mismatch, so an N-region layer
resolving N elements in each of M maps is withdrawn outright rather than
outlining the wrong country. The loss falls on the layer drawn on geo
alone: measured on a three-map figure, its region-1 selector matched 3
(USA, FRA, IND) while g.geo.geo2 and g.geo.geo10 matched 1 apiece, since
only the doubled id collapses to an ambiguous token. So the symptom is one
map on a figure silently losing its outline while its neighbours keep
theirs — which is also why nothing upstream flagged it.
The exact-attribute form is right for this family alone. Measured beside it,
a polar subplot's attribute is the bare id — class="polar",
class="polar2" — and a ternary's likewise, so g.polar already names one
subplot and g[class='polar polar'] matched nothing at all. That is why
polarSeriesSelectors scopes by token and is correct as written.
Choropleth selectors: one region at a time, in the order the trace declared them.
Plotly draws a path per calc entry — including the entries it resolved to no region at all, which stay in the DOM with no shape — so a region is addressed by its own position rather than by counting the ones before it. That keeps the pairing right for a map whose data has holes, which is the ordinary case: a value the source had nothing for is still a row.
The trace's group is counted within its own geo subplot, since plotly hangs no uid class on these groups either.
The subplot is named by its whole class attribute, not by a class token. Plotly writes a geo subplot's attribute as
"geo " + id, so the first one readsclass="geo geo"— a doubled string whoseclassListis the single token["geo"], which is a token every other geo subplot carries as well (class="geo geo2"→["geo", "geo2"]). Measured in Chromium on a two-map figure:.geolayer > g.geoand.geolayer > g.geo.geoare the same selector and both matched 2 of 2 subplots, so every region selector scoped togeoresolved to one path in every map — region 1 came back as USA and FRA at once.g[class='geo geo']matched exactly 1, and attribute matching is whole-string, sogeo10collides with neithergeonorgeo2.What that cost was the first map's whole highlight.
ChoroplethTrace.mapToSvgElementscompares the resolved elements against the declared regions and returns null on any mismatch, so an N-region layer resolving N elements in each of M maps is withdrawn outright rather than outlining the wrong country. The loss falls on the layer drawn ongeoalone: measured on a three-map figure, its region-1 selector matched 3 (USA, FRA, IND) whileg.geo.geo2andg.geo.geo10matched 1 apiece, since only the doubled id collapses to an ambiguous token. So the symptom is one map on a figure silently losing its outline while its neighbours keep theirs — which is also why nothing upstream flagged it.The exact-attribute form is right for this family alone. Measured beside it, a polar subplot's attribute is the bare id —
class="polar",class="polar2"— and a ternary's likewise, sog.polaralready names one subplot andg[class='polar polar']matched nothing at all. That is why polarSeriesSelectors scopes by token and is correct as written.