Ambiguous Optional Parentheses, Part One

(This is part one of a three-part series on C# language design issues involving elided parentheses in constructor calls. Part two is here.)

Another amusing question from StackOverflow, which I present here as a dialogue, as is my wont:

In C# 3.0 the object initializer (and collection initializer) syntax permits both new Foo { Bar = 123 } and new Foo() { Bar = 123 }. If the default constructor is used then the argument list is optional. Why did the design committee decide to make the argument list optional?

That decision was made in 2004, before I started on the C# team, so any reasoning about it will be somewhat conjectural. In browsing the archive of design notes there is no place where this decision is specifically documented with pros and cons, but I did discover a few things in looking at the notes.

First, even back when the idea was first raised, the sample code fragments in the notes for possible syntaxes do not include empty parameter lists. It seems that people thought this was the natural way to go from the beginning.

Second, I was also interested to discover that the design team was not entirely happy at first with the proposed syntax (of following the constructor call with a parenthesized list of side effects) but several efforts to find a better syntax that did not have ambiguities with existing programs failed to come up with anything better.

And third, it seems pretty natural to omit the arguments when initializing an object or collection because that's how arrays do it too: new int[] {1, 2, 3} and new List<int> {1, 2, 3} are obviously analogues.

OK, then, more generally: how do you decide when it is acceptable to make this sort of minor syntactic sugar?

The first question must be "is there a good reason why the argument list is required?" We cannot make it optional if it is required, obviously. In this case there is no compelling reason why it should be required. An empty argument list is redundant and adds no value, so if it can be elided, that's sort of nice. Furthermore, the feature hits a bit of a "sweet spot"; the most likely scenario for using an object or collection initializer is precisely when a mutable object has properties that are not set by any constructor, and such objects often have a default constructor. It seems reasonable that most of the time the argument list is going to be empty, so again, this feature is nice.

However, "sort of nice" only cuts it if the costs are low.

What costs do you consider?

We consider:

Design costs: is it going to take a lot of time to design the feature, verify that the design is sound, and write a specification? This is a tiny part of a larger feature; the cost is pretty low.

Compiler development costs: in this case we were going to have to completely rewrite the "object creation expression" parser logic anyway; making the argument list optional was a tiny fraction of the cost of the larger feature.

IDE development costs:: Does the proposed feature make it difficult to analyze an incomplete program as you are typing it? The IDE has to do that; features which make the IDE have to work much harder are bad features because the IDE has to be really snappy; it has to do semantic analysis between keystrokes. This sugar doesn't seem to add any complex lexical, grammatical or semantic analysis ambiguities. It seems to work well with refactorings, and other complex analysis that the IDE has to do.

Testing costs: does the feature introduce lots of scenarios that are difficult to test? In this case, no. It's very easy to test this sort of optional syntax, and again, the test cost of this part of the feature was a small fraction of the cost of testing the whole thing.

User Education costs: does the feature introduce syntax that will require a lot of documentation? Are we going to have to produce videos and blog posts about it? Are we going to get support calls from customers who are confused by the feature? For this sugar, no, no, and no.

Maintenance costs: what will the "bug tail" look like? Do we anticipate that we're likely to miss some crazy interaction with some other feature that causes a hard-to-reproduce and hard-to-fix bug? Or is it likely that we'll get it right the first time? In this case I do not recall there being any bugs reported against the optional-ness of the argument list. It's easy to get right.

Opportunity costs: by doing this sugar, are we spending lots of time and effort that we could be spending on something better for customers? Not really; again, this was a small fraction of the cost of the whole object-and-collection-initializers feature.

Future costs: Does the proposed feature "cut off" any anticipated future features, or make them more expensive and difficult? No. It seems unlikely that we're going to say in the future "well, we could have added metaprogramming to C# if only we had not made argument lists optional."

All the costs are low and the feature is nice to have, so we did the feature.

Why did you not add even more sugar at the same time?

Next time!

(This is part one of a three-part series on C# language design issues involving elided parentheses in constructor calls. Part two is here.)

Comments (9)

  1. James Dunne says:

    Nice post! Thanks for taking the time out of your busy schedule to answer such amusing questions :).

    My concern I tried to point out with the SO question was the inconsistency in constructor syntax this introduces and just to gauge why this inconsistency was introduced by the design team. Conditional optionality in syntax seems strange to me. That's all.

  2. Daniel Smith says:

    I didn't realise you could omit the parentheses like that.  It turns out you can also omit the initialiser contents within the curly brackets, and write weirdness such as:

    Foo foo = new Foo{}; // For those who don't like parentheses

  3. LBushkin says:

    I'm certainly curious about the last question you raise. One can certainly imagine a variety of syntactic sugar that could be added (like omitting the class name entirely in some cases of the initializer syntax)

    Foo f = { Bar = 5 };


    List<int> list = { 1, 2, 3, 4, 5 };

    Now with these, I suspect, there is far more potential for ambiguities and their clarity isn't great. I am definitely curious what other syntactic improvements were considered, but didn't make the cut.

  4. Joren says:

    "well, we could have added metaprogramming to C# (…)"

    You tease. 🙂

  5. Dave M says:


    Foo f = { Bar = 5 };

    The { Bar = 5 } syntax is used for anonymous types, but this statement has a named type. Therefore, this sugar would have the potential for confusion.

    Similarly, { 1, 2, 3, 4, 5 } defines an array. Using it to instantiate lists could also be seen as confusing.

    Not to mention that the 'new' keyword makes it obvious it's a reference type, and there's always the 'var' keyword for brevity and readability 😉

  6. LBushkin says:

    @Dave M : I completely agree – and I pointed that out (in general) in my question. I do, however, have an interest in language design and I am interested in how a team (like the C# team) analyzes and prioritizes such features. C# is no doubt a very interesting case study, in that regard, since the language team has to balance many competing concerns. Balancing things brevity against clarity, expressiveness vs. complexity, performance vs. modularity – these are all no doubt extremely challenging … especially for a mainstream language with millions of users who all expect their existing code to continue to work even as the language evolves.

  7. LBushkin says:

    @Joren: Indeed :o)   I would love to see what shape metaprogramming could take in C# if it were implemented. That and AOP would both be exciting enhancements to the language IMHO.

  8. Vince says:

    I never realised all the steps involved in a simple syntax change/feature like that. The mind boggles thinking about larger language features.

  9. Dan Vanderboom says:

    Good choice.  It seems we could also do away with the parameter list when not using an initializer.

    var world = new Canvas;

Skip to main content