Importance of Semantics

semantics1 | /sɪˈmæntɪks/ | noun (functioning as sing)

  1. the branch of linguistics that deals with the study of meaning, changes in meaning, and the principles that govern the relationship between sentences or words and their meanings
  2. the study of the relationships between signs and symbols and what they represent
  3. (logic)
    • the study of interpretations of a formal theory
    • the study of the relationship between the structure of a theory and its subject matter
    • (of a formal theory) the principles that determine the truth or falsehood of sentences within the theory, and the references of its terms

Semantics is ever more important in programming.

Take JDK Stack for example. If you were providing a Stack class, you should make sure to expose only those methods that do not break the stack semantics – add and remove elements only from the top of the stack. The underlying data structure could be an array or linked list (depending on your need). Your Stack should not expose methods to add or remove elements from other indices although you should expose the facility of iterating the elements. Exposing methods that mutate the order of elements freely at indices other than top is a breach of semantics. You could still use such a data structure and you could make it work to your wish but don’t call it Stack.

Java/JDK is notorious for such poor semantics. Not just Stack but many others – Thread, obtaining an immutable List, ambiguous method names2 etc.

Let us think about obtaining that immutable list. Here is how you get an immutable list out of an existing immutable list:-

// list could have been created elsewhere or in place new ArrayList();
Collections.immutableList(list);

You might argue it does the job. However it is operating on broken or poor semantics. With a List in hand, how do you tell that it is immutable? Imagine that you are returning the immutable list from one of your methods. How would the caller know that the list is immutable and avoid modifying the list. Don’t tell me from Java doc. It is known for sure only at runtime when an UnsupportedOperationException is encountered in an attempt to modify the list. A better semantic is to have a separate interface/class that does not even provide mutator methods just like IReadOnlyList in .NET.

Let us extend on the above thought a bit further. Suppose you have to implement an API List getMeTheList(/*some input arguments*/). The API can return an empty list for some cases and a non-empty one for others; some business logic. It is not uncommon to declare a shared empty list EMPTY_LIST and use that for all empty list cases in the above method and other such methods too. We don’t want to be churning empty lists for no reason. Remember you don’t the garbage collector for free. So you would make that shared empty list immutable – Collections.immutableList(new ArrayList()); If the shared list is not immutable, any consumer code that modified the list would make it non-empty and all consumers would not be getting empty list anymore when we return EMPTY_LIST.

final class AppLib {

   private static final List EMPTY_LIST_REF  = Collections.unmodifiableList(new ArrayList<>());

   @SuppressWarnings("unchecked")
   public static <T> List<T> emptyList() {
      return EMPTY_LIST_REF;
   }
}

class YourAppCode {
   public static <T> List<T> getMeTheList(/*some input arguments*/) {
      if (some truth condition) {
         return AppLib.emptyList();
      }

      if (some other condition) {
         return an existing list; like one from cache etc.
      }

      final List<T> result = new ArrayList<>();

      // some logic that populates the list

      return result;
   }
}

I am sure you see the problem with above implementation – inconsistent return value although it is List in all cases. Think from the caller’s perspective. There is no way that the caller can deduce (from the method signature) to refrain from modifying the List returned (or rather prevent the caller in the first place).

You can go to great lengths in making such a contract work with a try-catch, size()2 checks and what not. By doing so, you are just working around the problem rather than standardizing the semantics of the API; if you are the author or have access to effect a change in the API.

In majority of cases, the calling code iterates/consumes and does not make modifications to the resultant list. However, the fact that the calling code can make modifications should not influence the API design with a false idea that it is convenient for the caller. That said, if the API is designed with the information beforehand that the client code will make modifications or such is the regular scenario, then the semantics change; API should be returning a mutable list. But the API should be consistent in its behavior. In other words, the API should be semantically meaningful. Think of an Equals implementation. Likewise, you should also consider the effect of allowing or skipping nulls in your resultant list.

Your API that returns a Collection should not be returning a List sometimes and Set at other times.**

As a publisher of the API, you should first decide the appropriate semantics. No forethought of semantics or alternatively a poor/broken one is ignorantly or knowingly planting bugs in code. I am sure you would have come across APIs such as getMeTheList or worse used the JDK Stack in ways that ignore the stack semantics.

There are several other factors that determine the integrity of the semantics of an API – name, consistency, thread safety to name a few. Did I say name? Don’t make the mistake of thinking What’s in a name? Suppose you name your API float getProductWeight(), which weight does it return – absolute weight of the product, with addons, with packaging? For instance, think of a cycle. If getProductWeight will return different values depending on some context that is not even controlled by the caller, it will not only wrek havoc in your code(-base) or the caller code(-base) but it will be hiding in plain sight, and you will be busy fixing parts of the application except getProductWeight.

Among problems that affect the sanctity and the integrity of large scale application code bases, API semantics or the absence thereof is one of the biggest contributors.


  1. Definition of semantics was borrowed from [Dictionary.com](www.dictionary.com] 
  2. size() sounds the method is going to return the number of bytes consumed by the Collection. One of the more appropriate names is Count
Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s