Bug With Anonymous Records?

[cwarden]

Discussed in #2156

^{Originally posted by cwarden December 13, 2021}
I'm working with records for the first time and running into an unexpected issue. Running the following program with souffle -v, souffle seems to get stuck in a loop of ResolveAnonymousRecordAliases and FoldAnonymousRecords.

.type Cell = [
   row: number,
   column: number
]

.decl cell_size(cell: Cell, size: number)
.decl number_bigger(cell: Cell, num: number)

cell_size([0, 1], 5).
cell_size([1, 1], 3).

number_bigger([r1, p1], n) :-
   cell_size([r1, p1], A),
   n = count : { cell_size([r2, p2], B),
   [r1, p1] != [r2, p2],
   A < B}.

.output number_bigger(IO=stdout, headers=true)

Is it a bug or am I doing something wrong?

The issue occurs as of 244cb07.

[b-scholz]

The endless loop in the AST-transformation pipeline is caused by FoldAnonymousRecords since it can only deal with anonymous records outside of aggregates. In the reported example, we have the anonymous record in the sub-clause of count.

[b-scholz]

I have provided the fix so that aggregates are rewritten as well. The example would be rewritten to:

type Cell = [
   row: number,
   column: number
]

.decl cell_size(cell: Cell, size: number)
.decl number_bigger(cell: Cell, num: number)

cell_size([0, 1], 5).
cell_size([1, 1], 3).

number_bigger([r1, p1], n) :-
   cell_size([r1, p1], A),
   n = count : { cell_size([r2, p2], B),
   r1!=r2, p1!=  p2,
   A < B}.

.output number_bigger(IO=stdout, headers=true)

However, there is still the problem that the aggregation fails.

[b-scholz]

The problem is that we don't have a simple aggregate (although it looks like a simple aggregate).
The transformed Datalog code is,

.type Cell = [row:number, column:number]
.decl cell_size(cell:Cell, size:number) 
.decl number_bigger(cell:Cell, num:number) 
cell_size([0,1],5).

cell_size([1,1],3).

number_bigger([r1,p1],@generator_0) :- 
   cell_size([r1,p1],A),
   @generator_0 = count : { cell_size([r2,p2],B),r1 != r2,p1 != p2,A < B }.
.output number_bigger(IO="stdout",attributeNames="cell	num",headers="true",name="number_bigger",operation="output",output-dir=".",params="{"records": {"Cell": {"arity": 2, "params": ["row", "column"]}}, "relation": {"arity": 2, "params": ["cell", "num"]}}",types="{"ADTs": {}, "records": {"r:Cell": {"arity": 2, "types": ["i:number", "i:number"]}}, "relation": {"arity": 2, "types": ["r:Cell", "i:number"]}}")

This will not work because we have an unpack inside the aggregate relation.
We would need to fix aggregates first.

[b-scholz]

You could introduce a helper relation that reflects the computations inside the count sub-clause. That might be the easiest way to solve your problem.

[cwarden]

I solved it with a relation that maps the Cell records to autoinc() values that I used in number_bigger, but a relation of (bigger: Cell, smaller: Cell) tuples works as well.