Function resolveFieldRef

Reads one declared fact off an author's row.

An explicit ref is used verbatim with no fallback: a name the author wrote that misses is their error, and reporting it — with warnUnresolvedRef, once the series is read — beats papering over it with a column they did not name. Only the defaulted path walks the chain, trying canonical first and then each alternative in order — the type's own chain from DECLARED_FIELD_REF_FALLBACKS where it has one, and FIELD_REF_FALLBACKS otherwise.

Every name, explicit or defaulted, is looked for in three places in turn: the row's own key, the dotted path it spells, and the row's properties — see readFromRow. Name priority dominates: properties.name beats a top-level id on a GeoJSON feature, because a region is called by its name rather than by its FIPS code.

undefined and null both mean "no ref given" here, exactly as they do in validateDeclaration: the slots this rides in are frequently untyped chart config, where null is how an author writes "not set", and reading row['null'] for one would be a puzzle with no explanation attached.

Returns undefined when nothing resolves, and the field is then omitted from the payload. Never substitute a zero, a false or a placeholder: a missing weight is a forest plot without weights, while a weight of 0 is a study that counted for nothing.

A falsy value that is present — 0, '', false, null — resolves and stops the walk. Only an undefined reading moves on to the next name.

An array counts as a row here rather than being rejected: a d3-hexbin bin is the array of points that fell in it, which is why length is one of count's fallbacks.