Refactor DataStore to use interned keys (InternedKey) instead of
string-based CellKey for O(1) hash and compare operations.
Introduce SymbolTable-backed interning for all category and item
names, storing them as Symbol identifiers throughout the data structure.
Add secondary index mapping (category, item) pairs to sets of interned
keys, enabling efficient partial match queries without scanning all cells.
Optimize matching_values() to avoid allocating CellKey strings by
working directly with interned keys and intersecting index sets.
Update all callers to use new API: iter_cells(), matching_values(),
and internal lookup_key() helper.
Co-Authored-By: fiddlerwoaroof/git-smart-commit (unsloth/Qwen3.5-35B-A3B-GGUF:Q5_K_M)
Categories on the None axis are excluded from the grid and cell keys.
When evaluating cells, values across hidden dimensions are aggregated
using a per-measure function (default SUM). Adds evaluate_aggregated
to Model, none_cats to GridLayout, and 'n' shortcut in TileSelect.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The two tests were previously silenced when item_by_name was removed.
Rewrites them using category.items.get() directly, restoring coverage
of item-name and item-group serialization round-trips.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Removes unused methods (sum_matching, get_mut, item_by_name, item_index,
top_level_groups, is_group_collapsed, show_item) and unused constants
(LABEL_THRESHOLD, MIN_COL_WIDTH).
The sum_matching tests in model.rs were bypassing the formula evaluator
entirely. Replaced them with equivalent tests that call evaluate() against
the existing Total = SUM(Revenue) formula, exercising the real aggregation
code path.
Also fixes a compile error in view.rs prop_tests where View/Axis imports
and a doc comment were incorrectly commented out.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Root cause: set_axis silently ignores unregistered categories, so a
view section appearing before its categories would produce wrong axis
assignments when on_category_added later ran and assigned defaults.
Fix: collect all raw data in pass 1, then build the model in the
correct dependency order in pass 2 (categories → views → data/formulas).
The file can now list sections in any order.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
New format is diffable plain text with categories, items, formulas,
data cells, and views all readable without tooling. Legacy JSON files
(detected by leading '{') still load correctly for backwards compat.
Format overview:
# Model Name
## Category: Type
- Food
- Gas [Essentials]
## Data
Month=Jan, Type=Food = 100
## View: Default (active)
Type: row
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The CSV export used Box<dyn Iterator<Item = Option<usize>>> to unify
empty and non-empty row iteration, violating the CLAUDE.md rule that
Box/Rc container management should be split from logic. Replaced with a
direct for loop over row indices, removing both the Box and the Option
sentinels used to represent "placeholder empty row/col".
Also removes unused pub use cell::CellKey re-export and an unused
import in cell.rs tests.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously CellValue had three variants: Number, Text, and Empty.
The Empty variant acted as a null sentinel, but the compiler could not
distinguish between "this is a real value" and "this might be empty".
Code that received a CellValue could use it without checking for Empty,
because there was no type-level enforcement.
Now CellValue has only Number and Text. The absence of a value is
represented as None at every API boundary:
DataStore::get() → Option<&CellValue> (was &CellValue / Empty)
Model::get_cell() → Option<&CellValue> (was &CellValue / Empty)
Model::evaluate() → Option<CellValue> (was CellValue::Empty)
eval_formula() → Option<CellValue> (was CellValue::Empty)
Model gains clear_cell() for explicit key removal; ClearCell dispatch
calls it instead of set_cell(key, CellValue::Empty).
The compiler now forces every caller of evaluate/get_cell to handle
the None case explicitly — accidental use of an empty value as if it
were real is caught at compile time rather than silently computing
wrong results.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Three copies of cross_product existed (grid.rs, app.rs, persistence/mod.rs)
with slightly different signatures. Extracted into GridLayout in
src/view/layout.rs, which is now the single canonical mapping from a View
to a 2-D grid: row/col counts, labels, and cell_key(row, col) → CellKey.
All consumers updated to use GridLayout::new(model, view):
- grid.rs: render_grid, total-row computation, page bar
- persistence/mod.rs: export_csv
- app.rs: move_selection, jump_to_last_row/col, scroll_rows,
search_navigate, selected_cell_key
Also includes two app.rs UI bug fixes that were discovered while
refactoring:
- Ctrl+Arrow tile movement was unreachable (shadowed by plain arrow arms);
moved before plain arrow handlers
- RemoveFormula dispatch now passes target_category (required by the
formula management fix in the previous commit)
GridLayout has 6 unit tests covering counts, label formatting, cell_key
correctness, out-of-bounds, page coord inclusion, and evaluate round-trip.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- grid.rs: honour view.number_format (",.0" default, ",.2", ".4", etc.)
via parse_number_format/format_f64(n,comma,decimals); format_f64 now
pub so callers can reuse the same formatting logic.
- app.rs: n/N actually navigate to next/prev search match (cross-product
aware); fix dead unreachable N arm; add :set-format / :fmt command to
change the active view's number_format at runtime.
- persistence/mod.rs: CSV export now uses full cross-product of all
row/col-axis categories, matching grid rendering behaviour.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
