tessera 0.2.2

Analyze, group and aggregate tabular data: a pivot-table engine with expandable row/column hierarchies. Pure Dart; tessera_flutter adds the widgets.

0.2.2

• ExportTable: a plain table — not a cube — for the new table exporters: ExportColumns (header, source name, type, numberFormat, width; showsTime decides date vs. date-time by the type) and rows of values. ExportTable.ofFacts(facts, columns:, rows:, filter:, useLabels:) reads a fact table lazily (e.g. the facts behind a cell, or those passing the cube's filter); withColumn adjusts one column. TableExportTheme is the format-neutral look (header fill and font, data font, borders) that tessera_xlsx's XlsxTableExporter and tessera_ods's OdsTableExporter take.
• CsvTableExporter: an ExportTable as CSV — header row, numbers plainly with the options' decimal mark, days as yyyy-MM-dd, points in time as ISO 8601 — so a fact table survives a CSV round trip with its types. CsvExportOptions.field quotes one field (what both CSV exporters use).
• JsonFactExporter: records, export, exportLines and writeLines take rows: and filter: like ExportTable.ofFacts.
• Schema.structureKey: a canonical key for the structure of a source (the JSON list of its column names, in order), so that applications can remember schema edits or layouts per structure and apply them to the next file of the same shape. Built from names only on purpose: inferred types vary with the sampled rows.
• TesseraSnapshot.mimeType (application/vnd.tessera.snapshot) and TesseraSnapshot.fileExtension (tsnp): the official media type and extension of snapshots, so that servers, file dialogs and file associations all declare the same value. Documented in docs/snapshot.md.

0.2.1

• JsonDataSource (a JSON array of objects, parsed whole) and JsonlDataSource (JSON Lines, streamed with a row estimate), with JsonOptions: nested objects flattened into dotted columns (or kept as JSON text), arrays as JSON text, columns to name the columns, scanAllRows to union every record's keys. Values keep their JSON types; strings go through the schema's parsers like CSV cells.
• JsonFactExporter: a FactTable as records — one object per fact with a field per column (columns to pick, useLabels for label keys) — as a JSON array (export, optionally indented) or JSON Lines (exportLines, writeLines streams row by row; records is lazy). Integers stay integers, dates become ISO text, so JsonlDataSource reads the table back with its types.
• The default date formats also recognize the full ISO 8601 forms JSON producers write: yyyy-MM-ddTHH:mm:ss.SSS, with or without a trailing Z.
• JsonCubeExporter: a layout as records — one object per grid row, the row dimensions as fields named after their titles, one field per column entry and aggregate named from the labels — as a JSON array (export, optionally indented) or JSON Lines (exportLines, writeLines); records gives the objects. JsonExportOptions: group labels repeated (default) or once per group, the path separator, the indent.
• Chart series: ChartData (categories × series of doubles) and ScatterData ((x, y) points in series) for feeding any chart library. fromLayout reads a layout's cells as displayed (ChartEntries selects the entries: visible leaves by default, one level, given paths, or all), fromFacts builds a one- or two-dimensional cube for the purpose, fromCell drills into the facts behind one cell by another dimension, ScatterData.ofFacts plots individual facts (two measures, optional series dimension, row set and limit). Categories carry the raw dimension value (DateTime, int, double …) and periodStart composes nested date parts for a time axis; withoutEmpty(), transposed(). Coordinate.toFilter() and cellFilter(layout, cell) give the filter behind a cell.
• Continuous integration: a GitHub Actions workflow (.github/workflows/ci.yml at the repository root) runs the analyzer, the format check and every package's tests on each push and pull request, with LibreOffice, poppler, qpdf, ghostscript, librsvg and Noto Sans installed so the exporter round-trip tests run rather than skip.

0.2.0

The expression language and everything built on it. One breaking change.

Breaking

• Measure is sealed: Measure('total') still constructs a ColumnMeasure, but Measure.column moved to ColumnMeasure; every measure has id, label, explicitLabel, labelFor and columns. Dimension gained sourceColumns. AxisSide now lives in the engine (tessera_flutter re-exports it, so no import change there).

Expression language

• Expression (parse, check, validate, compile), ExpressionScope (row or cell scope over a FactTable or Schema), ExpressionError (kind, source range, arguments), ExprType, FunctionRegistry / ExpressionFunction (built-in functions plus application ones), the Expr syntax tree with toSource. SQL/Excel-flavoured syntax (and, or, not, =, <>, in, between, is empty, if(), #2024-01-31# dates, [quoted names]), static typing, SQL three-valued null handling, day arithmetic on dates, compiled to closures over the typed column arrays. TesseraStrings.expressionError gives the message in every locale.
• Plug points: ExpressionFilter; Measure.expression (ExpressionMeasure, materialized once per fact table); ExpressionDimension; Aggregate.expression (ExpressionAggregate, a cell formula over sum(x), avg(x), count, count(x), min, max, stdev, stdevp, var, varp and distinct(x), where x is a column or any row expression — sum(qty * price) / sum(qty)), and the DerivedAggregate base for aggregates computed from other aggregates' results.

Aggregates

• Statistical: Aggregate.stdDev / stdDevPopulation / variance / variancePopulation (Excel's STDEV.S / STDEV.P / VAR.S / VAR.P), one Welford accumulator merged with Chan's formula so parents are exact and large means do not cancel.
• Layout-relative ("Show Values As"): Aggregate.percentOf (TotalOf.row / column / grand / parentRow / parentColumn), differenceFrom and percentDifferenceFrom (BaseItem.previous / next / value(v) along an AxisSide), runningTotal, rank; the LayoutAggregate base with LayoutCellContext for custom ones; ValueDisplay names the ready-made choices for pickers. Computed when a cell is read; sorting by one sorts by its base. CubeCell.aggregate also accepts the dependencies of a derived aggregate.
• AggregateKind covers the new kinds; every label is localized.

Filters

• Structured filters CompareFilter, RangeFilter, TextFilter, EmptyFilter over a column, equal by value like ValueFilter, AndFilter, OrFilter and NotFilter now are. FactFilter.toExpressionSource() on every filter (null where it cannot be expressed); FactFilter.compile binds a filter to a table once per scan.

Saving and transport

• CubeJson: JSON encoding and decoding of CubeConfig (a bundle of CubeSpec, both ExpansionStates and the Schema; CubeConfig.of(cube), toCube(facts)) and of every part on its own. JsonAdapter for custom aggregates, dimensions and filters; functions for decoded expressions; FormatException on malformed input, UnsupportedError for members without a JSON form.
• TesseraSnapshot: a FactTable (and optionally a CubeConfig) as one Uint8List — JSON header, raw little-endian column arrays — that decode loads back in a copy after validating every section. The format is specified in docs/snapshot.md for other writers. 2 M rows: 106 MB, ~40 ms to encode, ~60 ms to decode.

Strings

• New members for the filter editor, the aggregate picker's formula and label fields, "show values as", the statistical and layout-relative aggregate labels, and expression errors, in all fourteen locales. The count-of-facts function is "Count of records" in the picker (it said "rows" in most locales, which clashed with the cube's rows).

0.1.1

• README: the status note no longer says the package is unpublished.

0.1.0

• Initial release: data sources (CSV, lists), schema inference, the fact table importer, cubes with expandable row/column hierarchies, built-in aggregates, AxisGeometry for renderers, and TesseraStrings with fourteen built-in languages.
• Cube.expandRowLevel / collapseRowLevel / expandColumnLevel / collapseColumnLevel open or close every group of one level (keeping deeper expansions), rowsAddedByExpandingLevel / columnsAddedByExpandingLevel count the effect beforehand; ExpansionState.collapseLevel. Strings sortAscending, sortDescending, expandAll, collapseAll, inheritSort, totalsAtEnd, totalsAtStart, totalsHidden, subtotalsAbove, subtotalsBelow, subtotalsLeft, subtotalsRight, subtotalsHidden.
• CubeAxis.subtotalPosition (SubtotalPosition.top / bottom / hidden): where an expanded group's own row goes; AxisLayout.entryFor resolves groups without a row; HeaderArea.path.
• CubeGrid: a CubeLayout as a rectangular grid of typed cells with merged areas, the way CubeView shows it — what exporters render from.
• CubeExportTheme / ExportFont / NumberFormat: the format-neutral look of an exported document (fills, border, fonts per role, number format; brand(primary:), gradient, mix), shared by every exporter.
• Oklch (perceptual colour, to/from ARGB, gamut clipping) and HueLevels: nesting levels coloured by hue alone at one lightness, headers the same hue with more chroma, golden-angle steps so adding a level keeps the others' colours. CubeExportTheme.hueLevels, rowHeaderFills / columnHeaderFills per level, levelBasis (combined / row), fillOf / fontOf for exporters; GridCell.rowLevel / columnLevel.
• GridMetrics: pixel geometry of a CubeGrid (content-sized column widths from an estimate or a TextMeasurer, row heights from the fonts, offsets) for renderers that lay cells out themselves; GridPagination cuts it into pages with repeated header rows and columns; NumberFormat.format(value, strings).
• CsvCubeExporter (CsvExportOptions: delimiter, quote, line ending, decimal separator, BOM, group labels at the origin or repeated).
• FactTable.withLabels relabels columns without a re-import (data is shared).
• AxisDimension.sort is nullable: null inherits the ordering of the level above (CubeAxis.sortAt resolves it), so an aggregate sort on the first level orders every level.