Click here to Skip to main content
Click here to Skip to main content

A Money type for the CLR

, 18 Mar 2013
Rate this:
Please Sign up or sign in to vote.
A convenient, high-performance money structure for the CLR which handles arithmetic operations, currency types, formatting, and careful distribution and rounding without loss.

Introduction

A large number of the world's applications manipulate money values. Here is a convenient, high-performance money structure for the CLR which handles arithmetic operations, currency types, formatting, and careful distribution and rounding without loss.

Background

The CLR doesn't include a native money type. This could be seen as a shortcoming, or a reasonable design decision, given that it is an object-oriented framework, and adding your own type to encapsulate the data and behavior you need is pretty much its reason of being.

Money, however, is a type which is so primitive, and so pervasive, that its absence is notable. Other primitive types are built-in, after all. Martin Fowler considers this in Patterns of Enterprise Application Architecture [PEAA]:

A large proportion of the computers in this world manipulate money, so it's always puzzled me that money isn't actually a first class data type in any mainstream programming language. The lack of a type causes problems, the most obvious surrounding currencies. ... The good thing about object-oriented programming is that you can fix these problems by creating a Money class that handles them. Of course, it's still surprising that none of the mainstream base class libraries actually do this. (p. 488)

Using the Code

Using this Money type is easy: it works just like any other numeric type.

Money m1 = 1.25;
Money m2 = 0.75;

Money total = m1 + m2;
Money difference = m1 - m2;

You can create Money instances with a different currency than your current culture's currency:

Money m3 = new Money(1.25, Currency.Eur);

...but you can't perform operations on two instances with different currency types:

// throws an InvalidOperationException due to different currencies
total = m2 + m3;

To distribute money without losing fractions, use a MoneyDistributor:

Money amountToDistribute = 0.05M;

// two decimal places
MoneyDistributor distributor = new MoneyDistributor(amountToDistribute,
                                                    FractionReceivers.LastToFirst,
                                                    RoundingPlaces.Two);
Money[] distribution = distributor.Distribute(0.3M);

Assert.Equal(3, distribution.Length);
Assert.Equal(new Money(0.01M), distribution[0]);
Assert.Equal(new Money(0.02M), distribution[1]);
Assert.Equal(new Money(0.02M), distribution[2]);

That's all there is to using it! However, if you are interested in a good case-study of the inputs, rationale, and struggles of a type design, read on...

Analysis of Approaches to a Money Type for the CLR

Approach in this Implementation

Money

In [PEAA], Martin Fowler and Matt Foemmel's example shows Money as an object with value semantics. In the CLR, a value type is a first-class entity, and so it makes sense to combine these two and make Money a value type. This should also help deal with the issue of performance the authors bring up, since value types are not reference counted on the heap, meaning less pressure on the GC and therefore higher performance.

Also in [PEAA], the underlying type used to store the value was an Int64 (long). The authors then scale the value by some power of 10 (0 - 3) to represent fractional units. In this type, I kept the integral value storage, but opted to represent the fraction as a completely separate Int32, which is scaled by 10^9 (the largest power of 10 which fits into an Int32). This allows storing much smaller fractions, which is useful for intermediate computations. Due to this, the type is a fixed-decimal point numeric representation. This is how relational databases often store money, so it makes a natural fit. Another alternative was to represent the money value with a System.Decimal. The problem with this approach is that System.Decimal is a binary floating-point type, and binary floating-point types give us all sorts of headaches when computing with them and not treating the round-off or computational error accumulation with extreme care. These issues can be avoided by working with whole numbers and scaling to represent fractions.

I opted to separate the responsibility to allocate money in a defined distribution into another class: MoneyDistributor. The reasons for this are that I think it helps readability and conscientious use, and having both a divide operation and an allocate operation on the money class seemed to be a bit of a conflict of interest. By separating out the responsibility for not losing (or gaining!) fractions of money into a separate class, I force Money to admit that it can't really do a good job at dividing itself up, and rather entrusts this to another class. Further, this then allows subclassing of MoneyDistributor for custom behavior, which can no longer be done with Money, since it is a value type.

Currency

With regard to the currency aspect of money, the CLR doesn't have this as an available type. Java follows ISO 4217, and so it is pretty safe to add something like this - just follow the spec. My first inclination was to just make a Currency class with static fields: one for each currency. However, Jason Hunt's implementation used a CultureInfo instance to represent the currency, and it got me to think about how this is already somewhat present in the BCL. Since the CultureInfo classes are built around an Int32 identifier known as the LCID, it seems reasonable to use this as the currency identifier as well. However, after some observation, this turns out not to be a sound approach: not only are the culture-to-currency relationships not 1-to-1, but they are also not N-to-1, since some cultures use more than one currency. In the end, I used the ISO spec to generate some code to load lookup tables and keep everything related based on the ISO numeric code for any given currency. This code is an Int32, and this serves as the only field in a Currency instance, allowing me to represent the Currency as a value type as well. This makes serialization of a Money value quite simple: there are no reference fields in it! One last functionality of Currency to point out: IFormatProvider is implemented on it, so that when ToString() is called on Money, the associated Currency instance is passed with it and it gets formatted as expected.

Other Approaches

When I started working on this type, I didn't find an implementation of a money type for the CLR. Two things of note: the first is that I didn't quite search for the right keywords (apparently, no one calls the CLR by its name: "the CLR"; everyone refers to it in a roundabout way using "C#" or ".NET"), and the second is that a good implementation was published the same day completely independently, half way around the world! Let's compare these other implementations, with the goal of finding the best approach if one clearly exists.

Value Type or Reference Type Currency type Supports ISO 4217 Fixed or floating Handles distribution Arithmetic operations Handles formatting Currency instance formats the value Parses formatted strings
Jason Hunt's Reference CultureInfo / RegionInfo No Floating No Yes Yes* No No
Michael R. Brumm's Value Enum/ table lookup No Floating No No Yes* No Yes
chimeric69's Value CultureInfo / RegionInfo No Floating No Yes Yes No Yes
Pascal Lindelauf's Reference Custom reference type / Enum Yes Floating Yes Yes Yes* No No
This implem-entation Value Custom value type / table lookup Yes Fixed Yes Yes Yes Yes No
* (not as IFormattable)

Value vs. Reference Type

Given that value types are first-class in the CLR, it seems quite natural to use it as the basis for a money type. A given instance of Money is considered equal to another instance when the values are equal. This is known as having value semantics. Fowler calls types which have this characteristic "value-objects". On the other hand, you give up certain flexibility when defining it to be a value type vs. a reference type, namely: inheritance. This can get to be a problem if you want to keep your money allocation/distribution code with the money type and then modify it in a subclass, or if your database mapping layer depends on a reference type to wrap with database mapping code. I've handled the first concern by delegating the distribution responsibility to a separate class, and the second concern might indicate that it's time to get a more robust mapping layer. Given that a value type is intrinsically serializable, it shouldn't be hard at all to store this type in a database. There are some variations on how to do this depending on whether you are using a floating point or fixed point format for the value. More on this in "Fixed or floating".

Currency Type

The major decision here is to determine if a custom type is used, or a combination, or System.Globalization.CultureInfo/System.Globalization.RegionInfo to represent the currency, and, if a custom type is used, if it should be a full-fledged type or an enumeration. Although the CultureInfo/RegionInfo approach is immediately appealing, it leaves some ambiguities which are hard to resolve, since some regions use more than one currency, and various currencies are used in more than one region. The other drawback is that aspects of the ISO 4217 spec which covers currency are not directly accessible: notably the numeric code and the exponent. The custom type approach allows you to deal with this, but the enumeration approach leaves you with the need to lookup this information in a table or via some static accessor method. These could be encapsulated in the money type so users wouldn't need to handle the specifics outside of just the enumeration. The custom type approach seems to me to be the most sound, since there is a good amount of data to encapsulate (see "Supports ISO 4217"). The actual type could be a reference type, but since the data rarely changes, and is keyed by a numeric code, a value type with that code as the key into static lookup tables is efficient and allows the value to be embedded in the money type, making money values very portable. By building these static lookup tables from the ISO spec as well as enumerating CultureInfo/RegionInfo, you can have all the information you need about currency from a single Int32 encapsulated in a value type.

Supports ISO 4217

The world standard which governs currencies will have an impact on any Currency type design. Or at least it should. If CultureInfo/RegionInfo is used, this spec is indirectly observed. However, as pointed out in "Currency type", there are ambiguities when using CultureInfo/RegionInfo to represent currency, and implementing the spec fully in its own type resolves these. There are several components to each currency covered in the spec: currency name, a symbol, an exponent to indicate the smallest generally used division of the currency, a three letter code, and a numeric code. It is this extra information which serves as a major factor of how to represent Currency as a custom type, as noted more completely in "Currency type".

Handles Distribution Without Loss

Allocating funds over a number of divisions or distributions can lead to fractions of units being gained or lost, which can then be magnified by subsequent computation or storage and retrieval. People who handle money don't like this. In [PEAA], allocation is handled by picking the number of digits after the decimal to keep, and truncating the quotient of the division at that digit, then subtracting the sum of all quotients from the initial amount, and distributing the remainder among the quotients in smallest-decimal increments. The modification I make here from that text is the inclusion of the precision desired to truncate to (from 0 - 9 places after the decimal) as well as an enumeration of three methods of distribution: first-to-last, last-to-first, and random.

Fixed or Floating

Here's a question the answer to which sure to raise some eyebrows to those who haven't a firm grasp on the consequences of using floating-point types. If you've dealt with binary floating point numbers with fractions that cannot be represented exactly in base 2, but can be represented exactly in base 10, you're going to lose money, and probably a lot more than just the fractions of currency units that it appears to be at first. That's why, for a money type, it has to be a fixed format which is stored in base 10. By way of example, 1/3 is not exactly representable in base 10 or base 2. In base 10, a common way to split a unit three ways is to take the fractional units and give them to the last distribution, e.g., $1 three ways would be: $0.33, $0.33, and $0.34. Both 0.33 and 0.34 are representable in base 10; however, they are not representable exactly as a finite-length base 2 number, and some precision will be lost when doing so. This loss is often compounded or magnified during operations. There is really only one right answer here: fixed base 10.

Another benefit of choosing a fixed base 10 number: most relational databases have an exact numeric type which stores the value in base 10, so no loss will be incurred when storing and retrieving these values from a database. This will not always be true with base 2 numbers, and database mapping code will need to account for this if they are used instead. If your database doesn't have a base 10 number to store the money in, or it is inadequate, developing a custom strategy using 12 bytes (a long and an int) or 16 bytes with currency (a long and two ints) is straightforward.

Arithmetic Operations

In the CLR, you are able to define operators on a custom type. Given that money is involved in plenty of arithmetical operations, it is natural to do so. The only ones which should be suspected are multiplication and division, since using these to get distributions of money which need to add up to a specific total (or starting amount) without any loss or gain is not generally possible. [PEAA] uses an "allocate" method on the money type to replace division in these cases. A decision can be made on whether to keep this method with the money type or delegate it to a specialized class. The first method means that you don't need to know about a separate type, and most good IDEs will show you the allocate method and a keen developer should notice it, but that there is a bit of tension in the Money class on how and when to divide itself. And the second method means that you have a clearly defined responsibility, code which uses it standing out a bit more as something special going on, and the ability to subclass for custom behavior (assuming that Money is a value type), but the need to discover and learn about a second class. I prefer the separate class: it could be argued to be purely style, but if a distribution behavior change is needed (although when pressed, I couldn't admit to knowing when this would be needed), this approach would make it much cleaner.

Handles Formatting

The CLR has a pattern in place to support formatting of a type to a string representation: IFormattable. It makes good sense to support this interface and match developer expectations on how ToString() should work.

Currency Instance Formats Money Values

Following on supporting IFormattable, there is a standard formatting string to indicate that the value should be formatted as currency: "C". In order to support this universally, you can pass into IFormattable.ToString() or String.Format() a format provider which will be used to help format the type correctly. Using a currency instance as the IFormattable instance also seems natural. The only problem with this is a confusion which might arise when using a different currency than what the money value is expressed in: if the currency isn't checked and an exception thrown on a difference, a money value will be expressed in a different currency but with the original currency's numeric value. Some kind of currency conversion might need to be done instead, but this appears too complex and unwieldy to me.

Parses Formatted Strings

The inverse of formatting: parsing a string into an instance of money and related currency. No interface provides Parse() and (in 2.0 and on) TryParse(), but they are found as static methods on the BCL's primitive types. Seeing as we'd like to make our Money type as similar to BCL primitive types as we can, in order to create the illusion that it is part of the BCL, implementing these two methods becomes part of our job.

Source Update Notes

  • 2013-03-18: Updated the project to VS2012, implemented rounding on Money, implemented IComparible on Money, added Money extension methods to distribute without needing to create a MoneyDistributor instance, added TryParse static method on Money and Currency, added debugger visualization to Money

History

  • 2008-07-30: First version.
  • 2008-08-01: Source updated
  • 2013-03-18: Source updated

License

This article, along with any associated source code and files, is licensed under The Microsoft Public License (Ms-PL)

About the Author

codekaizen
Architect
United States United States
I'm a software engineer with 25 years of experience in areas from game and simulation development, enterprise development, systems management, machine learning, real-time and embedded systems development and geospaitial systems development.
 
You can find more of my work at http://www.codeplex.com and my articles at http://vectordotnet.blogspot.com/ and http://dotnoted.spaces.live.com.

Comments and Discussions

 
GeneralMy vote of 5 PinmemberArkitec18-Dec-13 8:26 
GeneralSystem.Decimal is not imprecise Pinmemberjfos25-Mar-13 8:50 
GeneralMy vote of 5 PinmemberShubham Choudhary20-Mar-13 2:58 
GeneralMy vote of 5 PinmemberWen Hao19-Mar-13 15:18 
GeneralMy vote of 5 PinmemberPrasad Khandekar18-Mar-13 22:17 
GeneralMy vote of 6 PinmemberŞafak Gür18-Mar-13 9:32 
QuestionIs this being mantained? PinmemberMario Valdeavellano6-Mar-13 13:58 
AnswerRe: Is this being mantained? Pinmembercodekaizen7-Mar-13 5:48 
GeneralRe: Is this being mantained? PinmemberŞafak Gür14-Mar-13 21:28 
GeneralRe: Is this being mantained? Pinmembercodekaizen18-Mar-13 7:07 
GeneralMy vote of 5 PinmembermasterJalchr28-Nov-10 10:12 
GeneralFormatting not working as expected PinmemberWill Young6-Dec-08 10:38 
Generalabout project Pinmemberhacker877-Aug-08 19:20 
hi i am amit from india studying in the coleege of the engg .......
i am in final year aand i am doing project int c# please guide me ...
GeneralRe: about project Pinmembercodekaizen7-Aug-08 19:43 
GeneralRe: about project PinmemberStrandedPirate29-Jul-12 9:28 
JokeMicheal Bolton's been there... PinmemberLandon Zabcik7-Aug-08 8:01 
GeneralRe: Micheal Bolton's been there... PinmemberAstraldust19-Mar-13 7:05 
QuestionMoneyDistributor Not Implemented? Pinmemberdjsdjsdjsdjs6-Aug-08 5:04 
AnswerRe: MoneyDistributor Not Implemented? Pinmembercodekaizen6-Aug-08 7:51 
GeneralRe: MoneyDistributor Not Implemented? PinmemberNorbert Haberl19-Sep-11 2:48 
QuestionDecimal? PinmemberEd.Poore2-Aug-08 5:32 
AnswerRe: Decimal? Pinmembercodekaizen2-Aug-08 6:44 
GeneralRe: Decimal? Pinmemberdgph17-Aug-09 22:12 
GeneralRe: Decimal? Pinmember0zkr19-Mar-13 19:27 
GeneralTests Failed PinmemberDoncp1-Aug-08 10:08 

General General    News News    Suggestion Suggestion    Question Question    Bug Bug    Answer Answer    Joke Joke    Rant Rant    Admin Admin   

Use Ctrl+Left/Right to switch messages, Ctrl+Up/Down to switch threads, Ctrl+Shift+Left/Right to switch pages.

| Advertise | Privacy | Mobile
Web01 | 2.8.140721.1 | Last Updated 18 Mar 2013
Article Copyright 2008 by codekaizen
Everything else Copyright © CodeProject, 1999-2014
Terms of Service
Layout: fixed | fluid