Concurrent Build Scheduler

Tasks are declared in a JSON dependency graph. The scheduler works out what can run now, runs those tasks in parallel, and unlocks the next ones as their dependencies complete.

The thread pool underneath is written by hand — worker threads, futures, locks and condition variables — because the point of the project was to build the scheduling machinery rather than call it.

Correct parallel scheduling is mostly about states and waiting. A task can be pending, ready, running, complete or cancelled, and the transition has to be atomic with respect to every worker asking "is there anything I can do".

Workers block on condition variables rather than polling, cancellation has to propagate without leaving a worker waiting on a task that will never finish, and resource limits mean a task that is ready is not always allowed to start.

• 01JSON-defined task dependency graph
• 02Executes shell and Python tasks
• 03Hand-written thread pool with worker threads
• 04Futures for task results
• 05Locks and condition variables for coordination
• 06Explicit task state management
• 07Task cancellation
• 08CPU and RAM resource allocation per task
• 09Parallel execution of independent tasks