“FDD” means “Failure Driven Design”.

FDD can prevent future bugs.

FDD is like “design rule” checking built into the dev app.

I spent hours tracking down a hoary bug in my logic.

I wrote:

… (Kind, Edge, edge)
… sourceof(Edge, Sender)
… targetof(Edge, Receiver)
… (direct_contains, Parent, Edge)

Which failed silently using PROLOG rules.

The final clause “… (direct_contains, Parent, Edge)” was not fulfufilled due to a bug in my logic - the outer-most container did not have a Parent.

Nothing could prevent me from making this mistake, but, it is possible to prevent other errors of this kind.

What I Wanted to Say

I wanted to say:

Forall “… (kind, Edge, edge)”, and, 
 There must be a fact “… sourceof(Edge, Sender)”, and,
 There must be a fact “… targetof(Edge, Receiver)”, and,
 There must be a fact “… (direct_contains, Parent, Edge)”
else
 fail loudly

What I Really Said

But, I wrote:

Forall “… (Kind, Edge, edge)”
 And, if there is a fact “… sourceof(Edge, Sender)”,
 And, if there is a fact “… targetof(Edge, Receiver)”,
 And, ifthere is a fact “… (direct_contains, Parent, Edge)”
 succeed

What’s the Difference?

A subtle difference.

The And version silently fails if ANY of the 4 clauses is false.

Subtle bugs are the hardest to find.

Preventing Errors Using FDD

Using FDD mentality, I can change the meaning of:

… (kind, Edge, edge)
… sourceof(Edge, Sender)
… targetof(Edge, Receiver)
… (direct_contains, Parent, Edge)

To mean what I wanted and simply push a button to regenerate the whole shebang, and, raise errors for any other latent bugs of this nature.

Example Design Rule Check

I can choose syntax to map to meaning - I might change the generated code to match what I wanted to say.

For example, I might generate code that matched the first clause, then asserted the rest of the clauses (with appropriate error messages), e.g.

… (kind, Edge, edge)
(… sourceof(Edge, Sender)” -> true ; format(user_error, “FAILURE sourceof(Edge,Sender) failed on ~w~n)”, [Edge])
(… (targetof(Edge, Receiver) -> true ; format(user_error, "FAILURE:targetof(Edge, Receiver) ~w~n)", [Edge])
(… (direct_contains, Parent, Edge)” -> true ; format(user_error, "FAILURE: direct_contains, Parent, Edge) ~w~n)", [Edge])

The above code looks horrible and hides the true essence of what I want to say, but, the machine doesn’t care and will check my other code for errors.

I’m using PROLOG (swipl) as assembler and I’m emitting assembler code (PROLOG) to do what I want.

I can modify the DSL syntax, push a button, and have ALL of my code checked.

I don’t need to look at the generated assembler code (PROLOG)¹.

Yes, there might be cases where I want to say AND instead of “there must be a fact”. I will modify (invent) the syntax to handle such cases. I expect that most of my code is based on “there must be a fact” rules, with AND rules being the exception. I intend to invent a syntax to express AND rules, and, I will use the FDD transpiler to find those cases for me (in the current code base).

FDD lets me make sweeping changes such as this.

I need to modify the transpiler, then I need to push a button (or invoke ./run.bash).

Not a lot of work, but it captures one of my pervasive errors in logic…

Yes, I could have written the code correctly in the first place, but that would require a lot more thinking on my part (and, probably, a PhD degree).

I said what I wanted to say. Now, using FDD, my only task is to make the computer (transpiler) do what I wanted…

FDD In Action