|
A version of my project is available at
https://gitlab.utwente.nl/s2516268/research-project-2023-1B. It contains a very simple fuzzer with many borrowed elements from the python fuzzer from xsmith-examples. I have replaced all the fresh-subtypes with
(fresh-type-variable integral-type fractional-type)instead,
which seems to cause no more issues. However, if this is a bug it would of course be interesting to my research.Thanks for the quick response, and I look forward to hearing from you. Everard de Vree Van: William G Hatch <william@hatch.uno>
Verzonden: dinsdag 16 januari 2024 05:05
Aan: Vree, E.A. de (Everard, Student B-TCS) <e.a.devree@student.utwente.nl> CC: xsmith-dev@flux.utah.edu <xsmith-dev@flux.utah.edu> Onderwerp: Re: [xsmith-dev] Numeric types not playing nice Hmm, that sounds like a bug. Can you share a link to your code?
This may be one of those corners of Xsmith that was built out but never seriously exercised. On Mon, Jan 15, 2024 at 12:39:31PM +0000, Vree, E.A. de (Everard, Student B-TCS) wrote: >Hello everyone, > >I am currently working on a Haskell fuzzer and new to Racket and Xsmith in general, so I apologize in advance if this question is trivial. There is something I don't understand about the types in Xsmith: >I emulate Haskell's numerical types like this: > >(define number-type (base-type 'number #:leaf? #f)) >(define real-type (base-type 'real number-type #:leaf? #f)) >(define integral-type (base-type 'integral real-type #:leaf? #f)) >(define int-type (base-type 'int integral-type)) >(define integer-type (base-type 'integer integral-type)) >(define fractional-type (base-type 'fractional real-type #:leaf? #f)) >(define float-type (base-type 'float fractional-type)) > >I then define the type info from the Minus _expression_ like this: > >(define numeric-bin-op-subtype (λ (n t) (hash 'l t 'r t))) > >[Minus [(fresh-subtype-of number-type) numeric-bin-op-subtype]] > >My assumption was that both Minus children nodes would be constrained to all the non-leaf nodes in either integral-type, or fractional-type. However, Xsmith still generates Minus nodes with 'l integral and 'r fractional, causing a type constraint violation in Haskell. I have since switched to using (fresh-type-variable integral-type fractional-type). This works, but I am still puzzled about fresh-subtypes. REPL experiments confirm that even fresh-subtypes of my real-types can be unified with both fractional-type and integral-type. What happens when a fresh-subtype is created? > >Regards, > >Everard de Vree |