/NimContracts

Nim Language design by contract (DbC) library.

Primary LanguageNimMIT LicenseMIT

Contracts

test nimble

This module is used to make contracts – elegant promises that pieces of code will fulfill certain conditions.

Contracts are essentially value counterparts of type-based concepts, generic bounds, type attributes (e.x. not nil) etc. While they work similarly to assertions, as they throw exceptions when broken, main reasons for using contracts include elegance of code, clarity of exception messages, intuitive semantics and ease of analysis, both for human and external tools.

As usage of this module isn't typical and some extra rules are applied, aside from documentation for each of module's elements the description of the module as a whole is included.

Where could I hear about it?

Contracts and static analysis were briefly mentioned in June 8th 2020 Nim blog post and Nim v1.2.0 changelog (it actually uses similar names, requires vs require). DrNim (by Araq) project uses such concepts too.

Hello contracts

For most of the features of this module, contractual macro should be used. As it generates exception-raising code, some exceptions should be imported too. For now, importing whole module is advised (at least until the submodules are introduced).

Example usage (hello world equivalent):

import contracts
from math import sqrt, floor
    
proc isqrt[T: SomeInteger](x: T): T {.contractual.} =
  require:
    x >= 0
  ensure:
    result * result <= x
    (result+1) * (result+1) > x
  body:
    (T)(x.toBiggestFloat().sqrt().floor().toBiggestInt())


echo isqrt(18)  # prints 4

echo isqrt(-8)  # runtime error:
                #   broke 'x >= 0' promised at FILE.nim(LINE,COLUMN)
                #   [PreConditionError]

Overview

It is advised to use contracts as part of a fine documentation. Even when disabled by a switch or pragma, they still look good and describe one's intentions well. Thanks to that, the code is often much easier to read even without comments. Also, a reader experienced in contractual design (and prefferably Nim) can read them faster than natural language as they are more compact.

Consider finding key's position in a sorted array:

 contractual:
   var a = 0
   var b = arr.len-1
   while a < b:
     invariant:
       if key in arr: key in arr[a..b]
       abs(a - b) < `abs(a - b)`
     ensure:
       if key in arr: key == arr[a]
     body:
       let mid = (a + b) div 2
       if arr[mid] < key:
         a = mid + 1
       else:
         b = mid

The first loop invariant is rather obvious. The other one is also intuitive if we know this module uses '`' character meaning roundly "previously" (in last iteration or yield, can also be used in postconditions to refer to values seen by preconditions). Actually, we could strengthen our invariants by adding information about how fast the range shrinks but it's not necessary to prove the algorithm works (although it is needed to prove how efficient it is).

While contracts can be of great help, they require care, especially when object-oriented behaviour is expected. For present version, one has to follow certain rules described in Object-Oriented Contracts section (in the future they will be followed automatically).

Contractual context

As already mentioned, for most of this module's features to actually work, the code has to be inside of a contractual block. For callables, it can be applied as a pragma.

Syntax:

 contractual:
   ...

 proc ... {.contractual.} = ...

There are several Contractual Contexts (CoCo) inside of contractual block. They are bound to plain Nim code blocks, so no additional marking is needed. CoCo include:

  • proc is procedure, converter or method declaration
  • loop is iterator declaration or either for or while loop
  • type is type declaration (to be used in future versions)

It is advised, although not needed, to use contractual block for whole modules and any modules that utilize them.

Context keywords

Keyword Related to Semantics proc loop type
require PreCondition is needed yes yes no
invariant Invariant doesn't change no yes yes
ensure PostCondition is provided yes yes no
body implementation how it is done yes yes no

Sections should be used in the same order as in the table above.

There is also promise keyword (related to CustomContract) which describes any conditions and as the only one doesn't require being used in contractual. Its usage discouraged if any other CoCo would do as its less descriptive and have no additional rules to it but for the assertion.

Other features

This module also includes a few non-context tools:

  • ghost makes a block of code active only when contracts are turned on
  • forall and forsome represent quantifiers
  • assume is used for assumptions for human reader and external tools (no code generated)
  • promise makes custom contracts

Documentation

Contracts can be treated as a part of the signature of the routine or type associated with it. Due to that, this module can generate additional code block for the entity's documentation so that the code would blend in with the signature.

Example code:

 proc isqrt[T: SomeInteger](x: T): T {.contractual.} =
   ## Integer square root function.
   require:
     x >= 0
   ensure:
     result >= 0
     result * result <= x
     (result+1) * (result+1) > x
   body:
     (T)(x.toBiggestFloat().sqrt().floor().toBiggestInt())

Generated docs:

 proc isqrt[T: SomeInteger](x: T): T
   requires:
     x >= 0
   ensures:
     result >= 0
     result * result <= x
     (result+1) * (result+1) > x

Integer square root function.

Doc comments can be added anywhere between contractual sections, they will be treated just like when written in one place.

Contracts docs generation can be disabled using noContractsDocs compilation flag.

Diagnostics

To diagnose contracts, use explainContracts compile flag. It provides diagnostic informations about contracts, according to its numerical value:

  • 0: None (default) --- doesn't show any diagnostics
  • 1: Output (when flag defined but has no value) --- show final code
  • 2: Basic --- show both source and final code
  • 3: Verbose --- as Basic + show each type of contract detected for the entity (directly or indirectly).

Non-obvious goodies

Nim has an undocumented problem with JS deepCopy implementation lacking. Contracts uses its own implementation of deepCopy for JS backend, thus solving the problem. Thanks to that, "previous values" can be used on JS backend just like on C or C++ ones.

Future

Exception awareness

As for now, contracts cannot refer to exceptions, although it is planned for future versions to enable such a feature. In core Nim, it is possible to inform about what types of exceptions can be throw in the callable. However, there is no possibility to express what conditions should be met for these exceptions to be thrown. Additionally, it should be possible to specify what conditions will be true if an exception raises (notice they are not necessarily the same conditions).

Object-Oriented Contracts

New context type is planned for future versions. With it's introduction both proc and loop contexts will be changed to force object-oriented rules. For now, user should follow them manually.

contract containing inheritance
type invariant added ?
pre-condition added alternative
loop invariant added added
post-condition added added

Comparison to other libraries

Nim built-in pragmas

Nim itself defines some contract pragmas, including requires, ensures, assume, assert, invariant. Note how many of them are very similar to those used by Contracts. However, these do not generate any runtime checks and are there only to be inspected by external tools. In comparison, Contracts actually generates runtime checks with meaningful errors.

If some tooling starts actually using these pragmas, Contracts might get an option to inject these pragmas too, so that external tooling can use them.

Contra

Contra is overall a much more minimalistic project when it comes to contracts, both feature-wise and user-experience-wise. Listing the differences briefly:

  • foolproofness
    • Contra asks the user to use particular code order (preconditions, then postconditions, then main code). Contracts actually enforces these rules.
    • Contra errors on import for unsuported targets. Contracts only errors if the user uses features that cannot be supported on the given target, and when it does, it generates a meaningful error message.
  • types of contracts
    • Contra only provides preconditions and postconditions for procs and iterators. Contracts also have loop and iterator invariants (and some plans for object and methods inheritance support).
    • Contra doesn't provide the "old/previous value" feature. Contracts does.
    • Contra can only handle unconditional contracts, i.e. returning bools. Contracts can also handle conditional contracts (functionally equivalent to if contract_cond: cond else: true, but semantically different).
  • documentation
    • Contra can only generate documentation for preconditions (and only in some cases). Contracts generate documentation for all signature-like contracts (preconditions, postconditions, invariants) by default.
  • errors
    • Contra uses AssertionError. Contracts uses a separate error type hierarchy with separate types for e.g. preconditions and postconditions. That way, the user can treat caller errors (i.e. breaking preconditions) differently from the provider errors (i.e. breaking postconditions).
    • Contra only shows the number of contract that has been broken and only stringifies the condition for preconditions. Contracts stringifies all the contracts in its error messages.
    • Contra only shows the location of the contract's code when abused. Contracts provides it for all contracts by default.
  • additional contract-related features
    • Contra doesn't have "ghost code" feature (i.e. code that is only generated if runtime contracts checks are enabled), although the user can replicate it with a when defined(contracts): statement. Contracts does and actually implements ghost in a similar way.
    • Contra doesn't define any contract-related helper functions, such as forall and forsome (although sequtils.allIt and sequtils.anyIt can be used). Contracts does.
  • other features
    • Contra defines some debug printing optimizations with term rewriting. Contracts does not, it's a library purely for design-by-contract.
    • Contra defines some assert-and-print-generated-code funcionality. Contracts does not, it's a high-level library.
    • Contra defines some immutability-related tools. Contracts does not, it's programming style-agnostic.
  • support
    • Both Contra and Contracts fully suppor compile-time execution ({.compiletime.} and static)
    • Contra seems to have problems for JavaScript target on Nim >1.0. Contracts supports all of it features on JavaScript target.
    • Contra seems to have problems for NimScript target on Nim > 1.0. Contracts supports its basic features on NimScript target ("old/previous values" feature cannot be used though).

FAQ

  • What about assume blocks?

    • Contra: Assume blocks produce no code at all and are only meant for human reading only, you can do that using discard or similar contruct on Nim. KISS.
    • Contracts: False. assume blocks are for both humans and static analysis tools. They are also checked semantically and type-wise, unlike comments or strings, but generate no code whatsoever, unlike actual code put in discard (which can but doesn't have to be optimised away). What's more, no static analysis tool can just guess than some string or discard is assumed to represent an assumption. Unless it's marked by a variable's name or a comment, human readers can't always do that either.
  • What about body blocks?

    • Contra: does NOT uses nor needs body blocks.
    • Contracts: does not in fact "need" them. They are used mostly for aesthetic reasons and user experience of users coming from languages that use them, like Ada or Cobra. It may even get optional in future releases.
  • What about invariant blocks?

    • Contra: You can pass Invariants on the postconditions block.
    • Contracts: False. Invariants only make sense for loops and iterators (for procs they're essentially just copying the same condition as both precondition and postcondition). Both libraries implement postconditions with a defer statement, which means it only runs after the return, not after each yield, like an invariant is supposed to.
  • What about forall and forsome blocks?

    • Contra: Use sequtils.filterIt, sequtils.mapIt, sequtils.keepItIf, sequtils.allIt, sequtils.anyIt, etc.
    • Contracts: True. forall and forsome are implementation-wise very close to sequtils.allIt and sequtils.anyIt (even their documentation say so). They were created first and foremost to improve readability, because Contracts assumes the more readable and self-explanatory the better, especially considering it's part of the docs. It also improves user experience for users coming from language which have such features, e.g. Ada and Cobra.
  • What about ghost block?

    • Contra: Use when defined(release): or when defined(contracts):
    • Contracts: True. ghost is actually implemented as when ghostly():. Introducing it serves two main reasons: to make it usable as a pragma and to improve user experience for users coming from Ada, which supports ghost.
  • Whats the performance and speed cost of using Contra and Contracts?

    • Contra: Zero cost at runtime, since it produces no code at all when build for Release.
    • Contracts: True for both, the approach is very similar.
  • How to use Contra and Contracts at Compile Time?

    • Contra: Add {.compiletime.} or static:.
    • Contracts: True for both. Note that {.compiletime.} procs cannot be used outside of other {.compiletime.}, macros and static blocks, because they do not exist at runtime. Using them elsewhere triggers a not-so-obvious error request to generate code for .compileTime proc.