The aggregations under key, each {column, function, name}, the column
left out for a function that takes none (PLAN §11.5), and a window
function's offset (Lag, Lead) or buckets (NTile), JSON numbers (PLAN
§11.9). Texts and numbers are kept exactly, '' included, for validation
to judge: an unknown, empty or misplaced function or setting is held, not
dropped (Q4), since an invalid node can't run. A name left out is read as
the auto-name for the use (Q3), or '' when there is none, and written
back; a setting left out stays out, which validation reports, as a Limit's
cleared size: never its default, so an invalid one is never saved as valid. An entry key this version doesn't
know could change the rows, so it makes the node unreadable, once every
entry has been read. Shared by the nodes that aggregate (Group, Partition).
The aggregations under
key, each{column, function, name}, the column left out for a function that takes none (PLAN §11.5), and a window function'soffset(Lag, Lead) orbuckets(NTile), JSON numbers (PLAN §11.9). Texts and numbers are kept exactly,''included, for validation to judge: an unknown, empty or misplaced function or setting is held, not dropped (Q4), since an invalid node can't run. A name left out is read as the auto-name for the use (Q3), or''when there is none, and written back; a setting left out stays out, which validation reports, as a Limit's cleared size: never its default, so an invalid one is never saved as valid. An entry key this version doesn't know could change the rows, so it makes the node unreadable, once every entry has been read. Shared by the nodes that aggregate (Group, Partition).