datagrok-selection

The active DataFrame is t; row selection lives at t.selection (a DG.BitSet).

The number-one trap: t.selection.setAll(false) clears the selection. setAll(true) selects every row. That is the opposite of filter, where setAll(true) is the cleared / no-filter state.

Quick reference

Need	Code
select all rows	t.selection.setAll(true)
clear the selection	t.selection.setAll(false)
invert	t.selection.invert()
select by predicate (replace)	t.selection.init(i => pred(i))
select by index list (replace)	t.selection.setAll(false, false); for (const i of idx) t.selection.set(i,true,false); t.selection.fireChanged();
add by predicate (union)	t.selection.or(DG.BitSet.create(t.rowCount, i => pred(i)))
remove indices	for (const i of idx) t.selection.set(i,false,false); t.selection.fireChanged();
intersect with predicate	t.selection.and(DG.BitSet.create(t.rowCount, i => pred(i)))
count selected	t.selection.trueCount
indices of selected rows	t.selection.getSelectedIndexes() (returns Int32Array)
is row i selected?	t.selection.get(i)
any row selected?	t.selection.anyTrue
selected rows as a new DF	t.clone(t.selection)
filter → selection (select visible)	t.selection.copyFrom(t.filter)
selection → filter (hide unselected)	t.filter.copyFrom(t.selection)
move current row	t.currentRowIdx = i (validate 0 <= i < t.rowCount)
current row object / cell	t.currentRow, t.currentCell
mouse-over row	t.mouseOverRowIdx
subscribe to selection changes	view.subs.push(t.onSelectionChanged.subscribe(() => …))

The polarity table — opposite of filter

Selection and filter use the same DG.BitSet class, but the meaning of true differs, so "clear" goes in opposite directions:

Action	Filter (df.filter)	Selection (df.selection)
clear (return to default)	setAll(true) (show all)	setAll(false) (none selected)
meaning of bit === true	row passes (visible)	row is selected
meaning of bit === false	row hidden	row not selected

The pair to memorize, side by side:

t.filter.setAll(true);      // clear filter    → every row visible
t.selection.setAll(false);  // clear selection → none selected

Predicate polarity matches across skills: in both t.filter.init(pred) and t.selection.init(pred), returning true includes the row. Only the clear idiom flips.

Selection vs current row vs mouse-over

Three independent row pointers — none is "the selected row" in the singular:

Concept	API	Cardinality	Event
Selection	t.selection (BitSet)	many rows (set)	t.onSelectionChanged
Current row	t.currentRowIdx / t.currentRow	one row (focus)	t.onCurrentRowChanged
Mouse-over	t.mouseOverRowIdx	one row (transient)	t.onMouseOverRowChanged

Setting t.currentRowIdx = 5 does not select row 5; selecting rows does not move the current row. When a user says "highlight row 5", the intent is ambiguous — pick the path matching context ("scroll to" → current row; "mark" → selection) or surface the ambiguity.

Reading selection state

Need	Code
rows currently selected	t.selection.trueCount
rows not selected	t.selection.falseCount
total rows	t.selection.length (same as t.rowCount)
indices of selected rows	t.selection.getSelectedIndexes() → Int32Array
is row i selected?	t.selection.get(i)
any selected?	t.selection.anyTrue
any unselected?	t.selection.anyFalse
next / prev set bit	t.selection.findNext(i, true) / findPrev(i, true)

// How many rows are currently selected?
return {selected: t.selection.trueCount, unselected: t.selection.falseCount, total: t.rowCount};

getSelectedIndexes() caches its result. The cache invalidates on public mutators (set, init, setAll, and/or/xor/andNot, invert, copyFrom). If you mutate via getBuffer(), call fireChanged() to invalidate.

Programmatic selection — modes

The four modes map directly to BitSet operations:

Mode	BitSet op	Pattern
replace	init(pred) / copyFrom(other)	new selection replaces current
add	or(other)	union with current
remove	andNot(other)	subtract from current
intersect	and(other)	keep only rows in both

Predicate

t.selection.init(pred) is the fast, buffer-direct path for replace:

// Select rows where age > 40 (replace).
const age = t.getCol('age');
t.selection.init(i => (age.get(i)) > 40);
return {selected: t.selection.trueCount};

For add / remove / intersect with a predicate, build a fresh BitSet and combine:

// Add to the current selection (union, don't replace).
const act = t.getCol('activity');
t.selection.or(DG.BitSet.create(t.rowCount, i => (act.get(i)) > 7));

Index list

t.selection.set(i, x, notify) writes one bit. Pass notify=false in a loop, then call fireChanged() once — otherwise N writes fire N events:

// Select explicit indices (replace). Zero the buffer, set bits with notify=false,
// fire one change at the end.
const idx = [0, 3, 5, 7, 11];
t.selection.setAll(false, false);
for (const i of idx) t.selection.set(i, true, false);
t.selection.fireChanged();

Anti-pattern (laggy UI):

// BAD — fires N events.
for (const i of indices) t.selection.set(i, true);

From another BitSet

t.selection.copyFrom(other);     // replace
t.selection.or(other);           // add
t.selection.andNot(other);       // remove
t.selection.and(other);          // intersect

Clearing & inverting

// Deselect everything. setAll(FALSE), not TRUE.
t.selection.setAll(false);

// Flip the current selection.
t.selection.invert();

t.selection.setAll(true) selects every row — including rows currently hidden by the filter. For "select all currently-visible rows", copy from the filter (see below).

Selected rows as a new DataFrame

const subset = t.clone(t.selection);

Same shape as t.clone(t.filter) — different mask:

Idiom	Mask used	Source DF modified?
t.clone(t.filter)	t.filter	no — clone of visible rows
t.clone(t.selection)	t.selection	no — clone of selected rows

// Open selected rows as a new table view.
const subset = t.clone(t.selection);
grok.shell.addTableView(subset);

If the selection is empty, clone returns a zero-row DataFrame (no error). Guard with t.selection.anyTrue first for user-facing flows.

Cross-skill bridges — selection ↔ filter

Idiom	Effect	Phrasing it answers
t.filter.copyFrom(t.selection)	hide non-selected rows in place	"show only", "filter to the selection"
t.selection.copyFrom(t.filter)	mark every visible row as selected	"select all currently-visible rows"

Critical disambiguation:

• "select rows [by / where / top N]" → set t.selection via t.selection.init(pred) or index-list pattern.
• "extract" / "copy" / "open as new table" / "give me a new table" → t.clone(t.selection) (new DataFrame; original unchanged).

t.selection.copyFrom(t.filter) fires onSelectionChanged but does NOT fire a filter event; the reverse holds for t.filter.copyFrom(t.selection).

Current row

// Focus row 5 (the platform usually does this on click; rarely needed in scripts).
if (5 >= 0 && 5 < t.rowCount)
  t.currentRowIdx = 5;

t.currentRow (a Row proxy) and t.currentCell (a Cell) follow currentRowIdx — read them on demand. Setting currentRowIdx = -1 is the valid "no current row" state.

Describing the current selection

No built-in "describe selection" — hand-roll a structured report: count, capped index list, sample row from the first selected index.

// Describe the current selection.
const sel = t.selection;
const all = sel.getSelectedIndexes();
const MAX_INDEXES = 50;
const MAX_SAMPLE_COLS = 12;
const indexes = Array.from(all.slice(0, MAX_INDEXES));

const out = {
  count: sel.trueCount,
  total: t.rowCount,
  indexes,
  currentRowIdx: t.currentRowIdx,
  sample: undefined,
};

if (all.length > 0) {
  const firstIdx = all[0];
  const cols = t.columns;
  const sample = {};
  const colCap = Math.min(cols.length, MAX_SAMPLE_COLS);
  for (let c = 0; c < colCap; c++) {
    const col = cols.byIndex(c);
    sample[col.name] = col.get(firstIdx);
  }
  out.sample = sample;
}
return out;

Selection event lifecycle

One event — unlike filter:

Event	When
t.onSelectionChanged	Any mutation on t.selection (set, init, setAll, and/or/xor/andNot, invert, copyFrom). Same observable as t.selection.onChanged.

// Log selection count whenever it changes.
const sub = t.onSelectionChanged.subscribe(() => {
  console.log(`selected: ${t.selection.trueCount}`);
});
view.subs.push(sub);

In a viewer / widget, push onto viewer.subs[] so the subscription cancels on detach (DG.Widget does this automatically).

Coalescing. Every BitSet mutator takes notify: boolean = true. Pass false in a loop and call fireChanged() once at the end:

for (let i = 0; i < indices.length; i++)
  t.selection.set(indices[i], true, false);  // notify = false
t.selection.fireChanged();                   // one coalesced event

Common patterns

// Toggle selection on a single row.
const i = 5;
t.selection.set(i, !t.selection.get(i));

// Top N by a column — sort, slice, set.
const idx = Array.from(t.getSortedOrder(['activity'], [false])).slice(0, 5);
t.selection.setAll(false, false);
for (const i of idx) t.selection.set(i, true, false);
t.selection.fireChanged();

// "Select what I see" → narrow → "filter to what I just selected" pipeline.
t.selection.copyFrom(t.filter);                                                // visible → selection
const flag = t.getCol('flag');
t.selection.and(DG.BitSet.create(t.rowCount, i => flag.get(i) === 'ok'));      // narrow
t.filter.copyFrom(t.selection);                                                // push back to filter

Anti-patterns

1. t.selection.setAll(true) to "clear" the selection. That selects every row — the opposite of what filter's clear does. Use t.selection.setAll(false).
2. t.filter.setAll(false) to "clear" the filter. Hides every row. Use t.filter.setAll(true). The cross-skill pair must be memorized together.
3. t.rows.select(rowPred) for "replace selection" in perf-critical code. Functional, but iterates the RowList and per-bit set — materially slower than t.selection.init(pred), which is buffer-direct.
4. t.selection.init(...) to add a condition to the selection. init zeros the buffer first. To extend, OR in DG.BitSet.create(t.rowCount, pred).
5. Per-bit set in a loop without notify=false. N writes → N events → laggy UI. Always set(i, x, false) in the loop and fireChanged() at the end.
6. t.selection.setAll(true) when the user means "select all visible". That includes filtered-out rows. Use t.selection.copyFrom(t.filter).
7. Combining selections across DataFrames of different row counts. dfA.selection.and(dfB.selection) only makes sense if rows are aligned 1:1. Cross-table sync by key columns is grok.data.linkTables(...).
8. t.rows.removeWhereIdx(pred) when asked to "remove from the selection". That deletes rows from the DF entirely. To narrow the selection, use andNot or set(i, false, false) in a loop.

Out of scope

• Filter operations. t.filter lives in datagrok-filtering. Same BitSet API, opposite clear polarity — see the polarity table above.
• Viewer-driven interactive selection. Lasso, grid click, shift-extend, ctrl-toggle — dispatched via BitSet.handleClick(rowPred, mouseEvent, ...). Covered by datagrok-viewers.
• Cross-table selection sync. grok.data.linkTables(t1, t2, k1, k2, [SYNC_TYPE.SELECTION_TO_SELECTION, ...]) establishes long-lived sync.
• Generic DataFrame cloning. For "copy of this DF with these columns" (no selection involved), see datagrok-df-and-columns.