If you have a cyclic dependency in your project then you more likely have a DAG that’s not sufficiently broken down. That is if A and B seem to depend on each other you have some sub tasks in each that need to be pulled out. If this still fails to break the apparent cycle, you’ve got a problem.

Another thing to keep in mind is that there's a distinction between a system having cyclic dependencies (I need A & B working because they communicate with each other), and the task ordering. During development you can end up with a partially correct system. If A sends to B, and B sends to A but I only have one person working on both parts, then it's possible I'll end up in a state where A sends to B but not the reverse. That's fine during development. Yes, the system is "wrong" per the design or spec. But it's a partial system (like a half-built bridge). Your tasks will appear to have a cycle, but in reality you will break it by selecting one portion to complete first so in the end it remains (based on the chosen path) a DAG.

There shouldn't be any cycles in properly broken out plan, because the flow of time creates a causal directness; a true cycle would prevent any of the tasks involved from starting, which obviously doesn't happen in real life. Sometimes you may need to break things down a bit more.

And that's why I'm asking for a tool that lets me model projects as Directed Acyclic Graphs of tasks. Because all I seem to be getting in popular tools is a flat list with some categorizing, or a single level of subtasks at best. I want to be able to express directly that this subtask has sub-subtasks, and that this sub-subtask depends on completion of a subtask from some other task (without making it a dependency between root tasks).