Type Variables and Sort Checking about liquid-fixpoint

[nikivazou]

@ranjitjhala I believe that the problem here is that there is no checking that qualifiers are bool (or prop) values, neither at liquidHaskell not at fixpoint. Shouldn't we be checking that?

For example, if you instead had tv r as a refinement, sort-check will fail:

constant tv : func(1, [RGRef @(0); @(0) ])

constraint: 
  env [r: {v: RGRef a | [] }]
  grd true
  lhs {v : L a | []}
  rhs {v : L a | [ ? (tv([r])) ]}
  id 1 

[ranjitjhala]

Don't follow -- the example below has no qualifiers?

On Thursday, September 10, 2015, Niki Vazou [email protected]
wrote:

@ranjitjhala https://github.com/ranjitjhala I believe that the problem
here is that there is no checking that qualifiers are bool (or prop)
values, neither at liquidHaskell not at fixpoint. Shouldn't we be checking
that?

For example, if you instead had tv r as a refinement, sort-check will
fail:

constant tv : func(1, [RGRef @(0); @(0) ])

constraint:
env [r: {v: RGRef a | [] }]
grd true
lhs {v : L a | []}
rhs {v : L a | [ ? (tv([r])) ]}
id 1

—
Reply to this email directly or view it on GitHub
#98 (comment)
.

Ranjit.

[nikivazou]

It does not have qualifiers. But checking that tv r :: bool fails.

If we were checking that the body of the qualifier is a bool then the initial example would create a proper error, instead of crashing at Z3.

Is there a reason why we do not check well-formedness of qualifiers?

[ranjitjhala]

Well, the reason is just that the qualifiers are implicitly checked when
they are instantiated -- so they better be BOOL when you instantiate
them, or else they are not a valid instantiation. So I'm puzzled as to why
this should get past the instantiation. (This plus maybe some BAD reasons
due to our earlier hacky handling of polymorphism...)

On Thursday, September 10, 2015, Niki Vazou [email protected]
wrote:

It does not have qualifiers. But checking that tv r :: bool fails.

If we were checking that the body of the qualifier is a bool then the
initial example would create a proper error, instead of crashing at Z3.

Is there a reason why we do not check well-formedness of qualifiers?

—
Reply to this email directly or view it on GitHub
#98 (comment)
.

Ranjit.

[nikivazou]

Ok, I will check why it got through instantiation. Other than that, wouldn't it be better if we check well-formedness of qualifiers at liquidHaskell level, so that we provide better error messages there?

An error at instantiation will still happen at fixpoint, so we cannot provide good error messages.

[nikivazou]

Fix with #451

[ranjitjhala]

Thanks @nikivazou !!

On Sep 15, 2015, at 12:52 AM, Niki Vazou [email protected] wrote:

Closed #98.

—
Reply to this email directly or view it on GitHub.