Macaw.Data.Operations.Transforms.Ladder.Packed (macaw_data v0.13.0)
View SourceA ladder piece as one binary (spec/04 §4.5.1; macaw_data#38, #46): its segments as
fixed-width records, so a partial the caller's cache holds is a few words on its heap,
an off-heap binary beside it, and the merge reads buffers rather than terms. The Elixir
walk and the native kernel (Ladder.partial/4) answer the same bytes.
The packed form, little-endian throughout:
<<1, s_type, y_type, 0, segments::32, heads::binary-size(8 * segments), buckets::binary>>s_typeandy_type:0every point'ss(ory) an int64,1every one a float64 (both0with no bucket);- a head per segment in row order, 8 bytes:
<<0, 0::24, row::32>>a gap, its first row;<<1, 0::24, buckets::32>>a run, its count of buckets; - the runs' buckets in order, 88 bytes each:
<<b::signed-64>>, then its first, last, least and greatest points, each<<row::32, s::64, y::64>>.
The term form, <<2, external term format of the segments>>, holds a piece whose
points are not all of one number type, or past int64 (a log scale's integer bucket, a
macaw form's mixed readings): every value is kept exactly, as the walk answered it.
Summary
Types
A piece's segments, as Ladder reads them.
Types
@type segments() :: [run: [{integer(), tuple()}], gap: non_neg_integer()]
A piece's segments, as Ladder reads them.
Functions
@spec located(binary(), non_neg_integer()) :: list()
A piece's segments read as view v's: each gap {:gap, {v, row}} and each bucket
{b, {v, summary}}, as Ladder.runs/1 joins them.
A piece's segments as one binary: packed where its points share their types, else as a term.
A piece's segments, as pack/1 was given them.