On Mon, Feb 20, 2017 at 2:23 PM, 'John Clements' via Racket Users <racket-users@googlegroups.com> wrote: > >> On Feb 20, 2017, at 11:06, Sam Tobin-Hochstadt <sa...@cs.indiana.edu> wrote: >> >> I don't think `check-expect` is doing something that's >> module-relative. For example, this program: >> >> #lang racket >> (require test-engine/racket-tests) >> (struct x (a)) >> (check-expect (x 1) (x 1)) >> (run-tests) >> (display-results) >> >> fails its test. `equal?` in Racket checks structural equality in a >> deep way, which is indeed a feature not found in Java and Python, but >> I don't think it's any different than overriding .equals() everywhere >> in Java to perform structural traversal. > > I agree; my only point is that those writing unit tests probably want to > define (or have defined) two separate notions of equality; one for use in > tests, and one for use by clients. And that testing frameworks should > accommodate/encourage this.
Maybe that's true, but I haven't wanted to do that in my Racket code using either `check-expect` in class or `check-equal` in Rackunit (or at least, not enough to define a new abstraction for it). Have you tried making a testing framework to accommodate this better? Sam > > John > >> >> Sam >> >> On Mon, Feb 20, 2017 at 1:53 PM, 'John Clements' via Racket Users >> <racket-users@googlegroups.com> wrote: >>> >>>> On Feb 19, 2017, at 14:40, Matthias Felleisen <matth...@ccs.neu.edu> wrote: >>>> >>>> >>>>> On Feb 18, 2017, at 11:12 PM, 'John Clements' via Racket Users >>>>> <racket-users@googlegroups.com> wrote: >>>>> >>>>> (cc:ak) >>>>> >>>>> Okay, this sounds just crushingly obvious now that I say it, but >>>>> honestly, I don’t think it’s occurred to me: >>>>> >>>>> One reason that equality is such a nightmare in Java and Python (which >>>>> turns out to be JUST AS BAD), is that those of us that actually want to >>>>> write unit test cases want *intensional* equality, not extensional >>>>> equality—but not pointer-based equality. >>>>> >>>>> I just came across a nice example while working on code for first-year >>>>> students (thanks, Aaron!). >>>>> >>>>> Specifically, imagine an “array” class that contains a backing array and >>>>> a length, where the backing array might be arbitrarily longer than the >>>>> length: >>>>> >>>>> (struct arr (vec len)) >>>>> >>>>> So, for instance, an array with two elements in it might be represented >>>>> as either >>>>> >>>>> (arr (vector “apple" “horse” #f #f #f) 2) >>>>> >>>>> or as >>>>> >>>>> (arr (vector “apple” “horse” #f) 2) >>>>> >>>>> —they both represent the array of length 2 that has “apple” as the first >>>>> element, and “horse” as the second. >>>>> >>>>> If I’m providing this library to be used by others, I probably want >>>>> extensional equality; I don’t want users of my library to be able to >>>>> distinguish between the two. However, if I’m writing unit tests for my >>>>> library, I definitely *do* want to be able to distinguish between the >>>>> two, for instance in checking the behavior of arrays that fill up and >>>>> need to be resized. Moreover, pointer-based equality—the == of Java, and >>>>> the `is` of Python (IIUC)—is also largely useless for unit tests. >>>>> >>>>> It’s probably not terribly controversial to suggest that neither Python >>>>> nor Java were designed with unit testing in mind. Accordingly, I >>>>> shouldn’t be surprised to discover that they don’t provide a natural >>>>> notion of equality that fits well with unit tests. >>>>> >>>>> So, I have three questions: >>>>> >>>>> 1) Is there an existing term for what I would call “functional >>>>> extensional equality”—that is, Racket’s ‘equal?’ ? >>>>> 2) Does what I wrote make sense? >>>>> 3) Has this been written down somewhere else, as part of a larger work? >>>> >>>> >>>> John, >>>> >>>> what you point out is that extensional equality depends on the point of >>>> view. In your specific example, ‘arr’ from inside the module — >>>> specifically for non-exported functions — equal? must take into account >>>> the full vector. For a client module, equality is a notion of projection >>>> (elimination of array fields). In particular, I would expect your module >>>> to declare the structure opaque so that equal? does not inspect the full >>>> array. Additionally, I would expect you to export a function called >>>> arr-equal? so that clients can determine whether two instances are >>>> extensionally equal from their perspective (which is apparently needed in >>>> your context). >>>> >>>> If you were in a typed functional world, I’d use a sealed structure for >>>> the module or at least an existential type so that the outside world >>>> cannot inspect arr’s vec filed. This shows that equality is a >>>> type-dependent notion — just as Reynolds said it. >>>> >>>> In a typed OO world such as Java, you’d override .equal, which is public >>>> and thus accessible. But for internal, private comparisons, you’d have to >>>> step outside of Java’s equality predicate and use a method that inspects >>>> the full vec field. Here this is relatively easy because you are dealing >>>> with just two slots. >>>> >>>> In any case, I doubt equal? helps you here. Did I misunderstand anything? >>> >>> No, I don’t think you missed anything, but from the standpoint of those >>> teaching early classes in either Java or Python (a standpoint that I >>> understand you are generally and mercifully not compelled to adopt), it >>> seems that there’s a missing piece in unit testing frameworks: namely, the >>> default check-expect form in these languages should not be one that refers >>> to the notion of outside-of-module-extensional equality, but rather one >>> that refers to inside-of-module-extensional equality, as check-expect does. >>> I’d be tempted to call this notion of equality “testing equality,” but I’ve >>> been called out before on my barbarous neologisms. (Not sure this one would >>> be considered barbarous, at any rate.) Additionally, students should >>> probably not be required to implement this notion of equality by hand. >>> Sounds like a nice three-page position paper, probably not good enough for >>> ITICSE, maybe SIGSCE (sigh). >>> >>> John >>> >>> >>> >>> -- >>> You received this message because you are subscribed to the Google Groups >>> "Racket Users" group. >>> To unsubscribe from this group and stop receiving emails from it, send an >>> email to racket-users+unsubscr...@googlegroups.com. >>> For more options, visit https://groups.google.com/d/optout. > > > > -- > You received this message because you are subscribed to the Google Groups > "Racket Users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to racket-users+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/d/optout. -- You received this message because you are subscribed to the Google Groups "Racket Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to racket-users+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
