Lectures

We break down how to stop turning main into a “command centre” and split a console application into three layers: CLI (input/output and command parsing), domain (rules and validations via require), and storage (data storage). Using expense tracking as an example, we build the Expense model, storage, and service; add a command parser; show the data flow (mermaid); and cover typical mistakes that cause layer boundaries to “creep”.

We break down how to “bake” architecture into packages and directories in a Kotlin project: why you should sync package with the folder structure, how to use import as an indicator of broken boundaries, how to split responsibilities by layers ( app.cli, domain.model, domain.service, storage), how to fix the dependency direction and avoid turning utils into a “dumping ground.” There is a diagram in mermaid, a practical file layout, and a walkthrough of common mistakes.

We break down how to design interfaces in Kotlin as boundaries between layers: why the contract should be “narrow”, how to separate domain rules from storage via ExpenseRepository, how to move report formatting into ReportExporter, why to return List instead of MutableList, when to take a snapshot via toList(), and how the CLI wires dependencies via constructor injection without mixing roles ( println stays in the CLI).

We break down the practice of refactoring “without a big bang”: we keep compilation and behavior invariants, take short verifiable steps, introduce explicit types instead of “ubiquitous strings” ( ParsedCommand), move routing into CommandRouter, carefully move files across packages, and replace direct dependencies via an intermediary layer ( MutableList -> ExpenseRepository). Separately, we discuss how to use require()/ check() for explicit contracts and which mistakes most often break refactoring.

We break down “fixation v1” for a CLI application: introduce a single result contract via sealed interface CommandResult ( Ok/ Error), centralize output with render(), define command handling as handle(...) -> CommandResult, show a minimal pipeline readln() → parse → handle → render → println(), and pin down common mistakes that break layer boundaries.