Selector Synthesis

Writing selectors by hand is easy once you know the model; tricky when you don't. Selector synthesis lets you point at atoms you want and have CnD propose a selector expression that picks out exactly those atoms.

What it does

You tell CnD "I want these atoms" (or these pairs of atoms) by selecting them across one or more instances. The synthesizer searches the space of selector expressions — identifiers, set operations, joins, closures — until it finds one that matches every example you gave and nothing else. The result is a selector you can paste directly into a selector: field in your spec.

Under the hood, it's a CEGIS-style (Counter-Example Guided Inductive Synthesis) search over the selector grammar.

When to use it

• You want to color / align / group a set of atoms but don't know the relational expression that picks them out.
• You want the synthesizer to generalize: give examples across multiple instances and the proposed selector will be the one that fits all of them.
• You want to derive a binary selector (for orientation, inferredEdge, etc.) from example pairs.

Using it

1. Open the Synthesis panel from the sidebar in the Graph view.
2. Choose atom (unary) or pair (binary) selector.
3. Pick one or more instances to pull examples from.
4. Click the atoms (or drag to pair them) that the selector should match.
5. Click Synthesize. The panel shows the proposed expression and which atoms/pairs it matches.
6. If it looks right, copy the expression into your .cnd spec.

If the synthesizer can't find a matching expression, try:

• Adding or removing examples to disambiguate.
• Raising the maxDepth (expression complexity) setting.
• Checking that your examples are consistent — a contradictory set has no valid selector.

Limits

• Synthesis time grows with expression complexity and the number of identifiers in the model.
• Very complex relational patterns may fall outside the grammar the synthesizer explores — when that happens, write the selector by hand.