feat: Allow a Filter to be used as the Expression in a SelectColumn.

[cpwright]

No description provided.

[cpwright]

Fix #6364. (The engine side, does not cover the gRPC side which requires wider consensus.)

[devinrsmith]

LGTM. @rcaudy for final pass.

[rcaudy]

Prefer bulk advance?

Suggested change

	final RowSet.Iterator inputIt = inputRowSet.iterator();
	final RowSet.Iterator trueIt = filtered.iterator()) {
	long nextTrue = trueIt.hasNext() ? trueIt.nextLong() : -1;
	int offset = 0;
	while (nextTrue >= 0) {
	// the input iterator is a superset of the true iterator, so we can always find out what
	// the next value is without needing to check hasNext
	final long nextInput = inputIt.nextLong();
	final boolean found = nextInput == nextTrue;
	booleanDestination.set(offset++, found);
	if (found) {
	nextTrue = trueIt.hasNext() ? trueIt.nextLong() : -1;
	}
	}
	// fill everything else up with false, because nothing else can match
	booleanDestination.fillWithBoxedValue(offset, booleanDestination.size() - offset, false);
	final RowSequence.Iterator inputRows = inputRowSet.getRowSequenceIterator();
	final RowSet.Iterator trueRows = filtered.iterator()) {
	int offset = 0;
	while (trueRows.hasNext()) {
	final long nextTrue = trueRows.nextLong();
	// Find all the false rows between the last consumed input row and the next true row
	final int falsesSkipped = (int) inputRows.advanceAndGetPositionDistance(nextTrue + 1) - 1;
	if (falsesSkipped > 0) {
	booleanDestination.fillWithValue(offset, falsesSkipped, false);
	offset += falsesSkipped;
	}
	booleanDestination.set(offset++, true);
	}
	final int remainingFalses = booleanDestination.size() - offset;
	// Fill everything else up with false, because we've exhausted the trues
	if (remainingFalses > 0) {
	booleanDestination.fillWithValue(offset, remainingFalses, false);
	}

[cpwright]

Original code (density is the fraction of true values, uniformly distributed):

Density: 1.0000, Nanos: 11560028500, per cell=11.56
Density: 0.9999, Nanos: 13277620834, per cell=13.28
Density: 0.9990, Nanos: 14978527958, per cell=14.98
Density: 0.8725, Nanos: 21772909750, per cell=21.77
Density: 0.7500, Nanos: 28369796167, per cell=28.37
Density: 0.5000, Nanos: 36947907333, per cell=36.95
Density: 0.2500, Nanos: 24845209958, per cell=24.85
Density: 0.1250, Nanos: 17964264291, per cell=17.96
Density: 0.0010, Nanos: 11357967500, per cell=11.36
Density: 0.0001, Nanos: 10771053167, per cell=10.77

New code:

Density: 1.0000, Nanos: 11589953167, per cell=11.59
Density: 0.9999, Nanos: 13430658834, per cell=13.43
Density: 0.9990, Nanos: 15374023083, per cell=15.37
Density: 0.8725, Nanos: 22405263417, per cell=22.41
Density: 0.7500, Nanos: 28906296084, per cell=28.91
Density: 0.5000, Nanos: 36402417875, per cell=36.40
Density: 0.2500, Nanos: 24535559166, per cell=24.54
Density: 0.1250, Nanos: 17518296583, per cell=17.52
Density: 0.0010, Nanos: 10686682416, per cell=10.69
Density: 0.0001, Nanos: 10635898708, per cell=10.64

[cpwright]

I can swap in which ever implementation you prefer at this point.

[rcaudy]

I think I like the new code better, but there's no strong case for either.

[rcaudy]

.

[rcaudy]

Maybe this means that when we gather filters that are dependencies in QueryTable or WouldMatch, we should only consider refreshing filters.

[deephaven-internal]

Labels indicate documentation is required. Issues for documentation have been opened:

Community: deephaven/deephaven-docs-community#366