Erroneous "variable only occurs once" warning, depends on clause order

[langston-barrett]

On this file:

.decl bar(?aCtx: symbol, ?alloc: symbol)
.decl foo()
foo() :-
  bar(?c1, ?a1),
  bar(?c2, ?a2),
  ( ?a1 != ?a2
  ; ?c1 != ?c2
  ).
foo() :-
  bar(?c1, ?a1),
  bar(?c2, ?a2),
  ( ?c1 != ?c2
  ; ?a1 != ?a2
  ).

Souffle prints these warnings:

Warning: No rules/facts defined for relation bar in file test.dl at line 1
.decl bar(?aCtx: symbol, ?alloc: symbol)
------^----------------------------------
Warning: Variable ?c1 only occurs once in file test.dl at line 4
  bar(?c1, ?a1),
------^----------
Warning: Variable ?c2 only occurs once in file test.dl at line 5
  bar(?c2, ?a2),
------^----------
Warning: Variable ?a1 only occurs once in file test.dl at line 10
  bar(?c1, ?a1),
-----------^-----
Warning: Variable ?a2 only occurs once in file test.dl at line 11
  bar(?c2, ?a2),
-----------^-----

Note that for the first rule for foo, only variables c1 and c2 are mentioned, and for the second rule, only a1 and a2` are mentioned, despite all variables appearing the same number of times in both rules. I wouldn't expect to see this warning at all, given that each variable appears more than once.

Version:

Souffle: 78c2bfc(32bit Domains)
Copyright (c) 2016-19 The Souffle Developers.
Copyright (c) 2013-16 Oracle and/or its affiliates.

[b-scholz]

That is a problem with how Souffle is processing disjunctions. It expands the program rules to the following rules:

foo() :-
  bar(?c1, ?a1),
  bar(?c2, ?a2),
  ?a1 != ?a2.
foo() :-
  bar(?c1, ?a1),
  bar(?c2, ?a2),
 ?c1 != ?c2.
foo() :-
  bar(?c1, ?a1),
  bar(?c2, ?a2),
  ?c1 != ?c2.
foo() :-
  bar(?c1, ?a1),
  bar(?c2, ?a2),
  ?a1 != ?a2.

You can check with --show=transformed-datalog. This transformation is done in the ParserTransformer (way before the AST pipeline with its semantic checker).

Only if disjunctions (hopefully someday) become part of the AST we can suppress these error messages.

[b-scholz]

This is related to issue #1289 and #1409.

[quentin]

These warnings won't appear anymore since #2415