An Ostler in IT

Horses for Courses

A first take on contracts in Clojure

I guess I’ve always written pretty defensive code (trust nothing), insisting that e.g. passed arguments are as expected and needed. Anybody who’s seen my recent Ruby code will have found it littered with many, many assertions. Usually my methods check both the input arguments for validity and also end with a call to a final guard to ensure the returned value is also valid.

During my time writing Perl (rather a long time ago), I used my own variants of the standard carp, croak and confess functions heavily, and followed received wisdom to fail early and often.

After moving to Ruby as my primary language, my Perl style evolved into the implementation of mustbe method e.g. mustbe-hash-or-croak to ensure the argument is as expected. My mustbe methods are dynamically defined in a mixin using standard Ruby’s metaprogramming techniques and, although they commonly test whether the argument is an instance of a given class, they can apply (assert) arbitrary constraints.

In time I realised I’d fallen into a contracts style of programming. I’d not sure how or where I first came across contracts formalism but it may have been when reading about Eiffel’s Design by Contract.

Eiffel’s DBC asserts pre and post constraints on a routine’s execution; in my own informal way, I’d become strongly attracted to contract assertion, not only on entry to and exit from routines (methods, functions), but also in their bodies.

My recent excursions into writing Clojure has had me reaching for a way (technique) to apply contracts. Hence this post.

Contracts in Clojure

I first stumbled on contracts in Clojure whilst reading the superb Joy of Clojure (JoC) written by Michael Fogus (hereafter just Fogus) and Chris Houser.

Fogus had already blogged about Clojure’s direct support for the enforcement (assertion) of a function’s entry (pre) and exit (post) constraints. For example, the following function ensures the passed argument is a positive number and a map (hash) is returned.

1
2
3
4
5
6
7
(defn my-constrained-function
  "This function constrains its input and output
  It ensures the passed argument is a positive number and returned value a map"
  [n]
  {:pre [(instance? Number n) (> n 0]
   :post [(map? %)]}
  {:return n})

Note both :pre and :post apply an and-ed sequence of constraints; pre has two while post has one in the above.

Note also, the return value is available in a :post constraint using the % familiar.

To me, one of the most interesting, powerful and exciting capabilities of Clojure’s support for contracts is the ability (opportunity) to decouple the constraints from the function the former are applied to. Consider this example:

The call to make-a-red-car will succeed but the call to make-an-eco-car will fail because the fuel given (petrol) is not one of the acceptable values in the post condition.

The take away here is that the more specific constraints (i.e. red or eco) can be external to the base make-a-car function. JoC observes (section 7.1.5) on this aspect-like nature:

By pulling out the assertions into a wrapper function, we’ve detached some domain- specific requirements from a potentially globally useful function and isolated them in aspects (Laddad 2003). By detaching pre- and postconditions from the functions them- selves, you can mix in any implementation that you please, knowing that as long as it fulfills the contract (Meyer 1991), its interposition is transparent.

Contracts Libraries for Clojure

Fogus followed up the JoC’s treatment of constraints with a blog announcing his implementation of a contracts library for Clojure: Trammel.

Also in the frame has been Dmitri Naumov’s clojure-contracts.

Earlier this year Trammel and clojure-contracts were brought together into clojure.core.contracts (CCC) which the Github page warns is a work in progress. So, assuming the core library has a more assured future than the earlier two, this post explores using CCC. Documentation for CCC is less well developed than e.g. Trammel: I found myself looking at both Trammel’s examples and also CCC’s code to get my head fully around what was going on.

Exploring clojure.core.contracts

The core idea behind CCC is both simple and elegant. Contracts are applied to a target function by “wrapping” the latter in anonymous functions which apply the pre and post constraints. Multiple contracts are applied by recursive definitions of wrapper functions (so each wrapper sees all the input arguments and their signature must be appropriate).

Much of the heavy lifting of CCC is accomplished by Clojure macros.

Defining an aspect contract

The following code generates two contracts. The first insists on the argument being a map, and the second on the return value being a number:

Notes:

  • the contract macro generate an anonymous contract but still must be supplied with a name (e.g. aspect-suck-a-map-cx) and description (e.g. enforce a map input argument).

  • To use the anonymous contract subsequently, the result of the contract macro must be assigned to a symbol (e.g. aspect-suck-a-map)

  • in the input a map contract (aspect-suck-a-map), a single pre condition has been specified (map? m)

  • in the output a number contract (aspect-spit-a-number) , a single post condition has been specified (instance Number %)

    post conditions appear after the =>

    The return value is available in post conditions as %

  • the arity of the contracts must be compatible to whatever subsequent function(s) they will be applied to (see later for more on this)

Defining a constrained function

The with-constraints function applies 0, 1 or more aspect contracts to an existing function, creating a new function that enforces all the contracts. For example, in the following example contracts are used to ensure the input argument to the guaranteed-suck-a-map-and-spit-a-number function is definitely a map, and the return value definitely a number. The unconstrained function (suck-a-map-and-spit-a-number) offers no such guarantees.

Applying contracts to existing functions

A refinement of with-constraints is provide which “updates” an existing function with contracts. Under the covers, provide uses with-constraints together with alter-var-root to update the binding of the target function to the (new) function generated by with-constraints. For example:

Notes:

  • provide can take a 0, 1 or more vectors, in this example only one is supplied:

  • provide offers an easy way to add (or subtract) contracts to a existing function, without making any other changes to the code.

  • contracts are applied in the order given.

A note on contract signatures and arities

Even though an output contract may only be interested in the return value (%), it must still have a signature that is compatible with every other contract used in the same with-constraints call.

This militates somewhat against composing multiple argument functions using generic aspect contracts as the “composite” signature and arity of the composed function must be consistent with all of the pre and post conditions / contracts.

The following code illustrates the point:

It would help some, especially for pure output contracts, if CCC supported varargs i.e the ability to have optional arguments after the % in the signature (e.g. [a b & c]). This currently fails because the build-contract-body function does not notice the % and includes it in an apply. However a one line modification to that function seems to do the job with no untoward side effects:

1
      '?ARGS       (vec (list* (remove #(= "&" (name %)) args)))

Some of the below relies on the use of varargs and hence the modification.

Contracts Sugar

So far, so good. However, it would be cumbrous to define the equivalent rich (large) portfolio of contracts / constrained functions I use in Ruby using the available CCC functions and macros. More of an exercise in writing Clojure macros, it is straightforward to implement a few productivity aids.

Generating many contracts easily

The code below shows a macro (make-contracts) that does much of the leg-work to define new contracts. Like provide, it accepts 0, 1 or more vectors. Note, these examples are for one argument contracts.

Generating mustbe functions

My Ruby mustbe methods mostly take one argument and test it, usually to confirm whether it is an instance of a class or not.

The follow code includes a macro (make-mustbe-functions) to do the necessary, leveraging make-contracts to define the aspect contracts. Note, the “target” function is always identity.

Generating contract functions for inline use

Sometimes, in “open code”, you will also want to apply one or more contracts to an value of an arbitrary expression. Again, a macro (make-inline-contract-functions) can help with the legwork. Note this example require my minor mod to support varags.

Overloading pre and post conditions support

In order to satisfy a constraint in a :pre or :post condition, the test function needs only to be (return) true: You can write any old condition function that accepts the arguments, does any processing it wants to, and as long as it return true, the constraint will pass. :pre and :post therefore allow arbitrary logic to be applied at a function’s entry and exit, not just performing argument checking.

For example, you may want to know (e.g. println) the return value from a function whilst testing, but turn off the println when in production. Using provide to update the definition of a target function makes its trivially easy to enable and disable such pseudo-contracts, with no run-time performance penalty either (since provide is a compile-time macro).

Another interesting possibility would be to use pseudo-contracts to support out-of-band auditing a function’s usage, perhaps adding authorisation checks.

Creating a pseudo-contract with an arbitrary code body, and probably without any actual conditions, that can be used with with-constraints and / or provide is not possible with the contract macro. But it is possible to create a new macro, and hack the other macros and functions (build-contract-body and build-contract-fn-body) to do the necessary. It works but the code feels messy and should probably be a feature request, together with varargs support, to Fogus for comment.

Final Words

CCC is a great addition to the tools you can use to build quality Clojure code. That said, support for rich usage (portfolio) of contracts feels a bit primitive right now. But, as my own simple attempts have demonstrated, it wouldn’t take much to add some productivity features.