Family Tree - Case Study

Overview

An iOS app for mapping your family without feeling like you are operating database software. The tree is the main screen: you add relatives from the person you are looking at, and empty "ghost" slots show where a missing parent, partner, or child would go.

Real families are hard to draw. Remarriages, step-parents, adoptions, and partners who married in all have to land somewhere sensible, and an early version put a newly added father 482 points from his own son. So the project was rebuilt around a pure-Swift layout engine in its own package, tested without a simulator, with SwiftUI only drawing what the engine decides.

Status: the foundation and the structural rebuild are done. Now in progress: interaction refinements found on a phone, then the first release — your own tree instead of the built-in sample family, saved data, editing and undo, and accessibility.

Swift SwiftUI Swift Package Swift Testing iOS 26 Xcode

Problem & Solution

The Problem

Family tree apps feel like filling in database forms
Remarriages and step-families break simple tree layouts
Layout bugs only showed up by eye, in the simulator
One 1,640-line view file mixed drawing with family logic
Stored relationships drifted out of step with each other

The Solution

Add relatives from the tree itself, in human terms
A layout engine with explicit rules for partners and families
Golden layout tests that run in about a second
Family logic in a package that never imports SwiftUI
Store two relationships; derive everything else

Key Highlights

One Architectural Rule
Family logic never imports SwiftUI, and UI code never does family math; the root view went from 1,640 lines and twelve types to 93 lines

Two Stored Relationships
Only parent-child and partner are stored. Siblings, labels like "cousin", and family groupings are all derived by walking those two

Order, Then Coordinates
The engine decides each row's order before any position exists and never reorders afterwards, which fixed relatives landing next to strangers

Ghost Slots
Placeholders for missing relatives are placed by the layout pass in space the row reserved, not by searching for a gap, which once put one 1,914 points from its owner

Fast Tests
80 package tests for family rules and layout, including golden layouts of fixture families, run in about a second; 5 app tests cover the store and viewport

Written Plan
Twelve design documents, a decision log, and an ordered work plan with acceptance criteria for every task

Technical Approach

Moved the family model, rules, and layout into a separate Swift package, so the logic can be tested in about a second without building the app or starting a simulator
Removed the stored sibling relationship entirely, because a pairwise link cannot carry a transitive fact; a sibling is now a statement about a shared parent
Made slot rules shape what the app offers, never what it permits, so a step-parent can be added as a third parent
Kept partners side by side and a child's co-parents together, with no one placed between them, even when no partnership is recorded
Routed connector lines in the engine, so every line joins up and no two family bars share a height and a stretch of canvas
Computed layout once per change and handed views a prepared value, after finding the canvas recomputed per node on every scroll frame
Distinguished relationships by line shape, not colour alone, so the meaning never depends on colour
Checked for cycles across all ancestors, not just direct two-person loops
Wrote every rule above into the project guide as a verified defect, so it cannot quietly come back

What's Next

Phase 1: Your Own Tree

Start from an empty tree — name yourself, then add your first relatives — instead of the built-in sample family
Save data with SwiftData, in a shape that can sync through iCloud later, with file protection on
Edit and delete people, disconnect relationships, and undo with the system gesture
Full VoiceOver labels, a deliberate reading order, a list alternative to the canvas, and Dynamic Type at the largest sizes
Choose an App Store name

What This Demonstrates

Knowing When to Rebuild: Stopped adding features to a layout that could only be checked by eye, and rebuilt it as an engine with tests first
Domain Modelling: Reduced family relationships to two stored facts and derived the rest, removing a whole class of inconsistency
Separation of Concerns: Enforced a single boundary between logic and drawing, and kept files small enough to read in one sitting
Learning From Defects: Turned each real layout bug into a written rule and a test
Planned Delivery: Worked from an ordered plan with acceptance criteria, checked on a phone simulator, not just a passing build

A Family Tree App With a Layout Engine You Can Test