Muldis::D::Outdated::Conventions - Style and design guidelines for Muldis D


This document is Muldis::D::Outdated::Conventions version 0.148.2.


This document is part of the Muldis D language specification, whose root document is Muldis::D; you should read that root document before you read this one, which provides subservient details.


This document briefly outlines various conventions used in the Muldis D language, and provides some style and design recommendations for code and/or projects written in Muldis D, or for alternate language dialects, or extensions, or implementations.

Unlike the rest of the Muldis D language specification, following anything stated herein is not necessary for conformation to the spec, and you should feel free to break these suggestions whenever it makes sense, just like when using any style rules.

This document is very much an early draft, and for not just contains a bit of brainstorming in point form.


Entity Naming

Note that any use of the word "name" specifically refers to a Name, or in other words the declared unqualified name of an entity, which sometimes is always used in that form, and which other times forms an element of a NameChain.

Although Muldis D allows names to be comprised of any characters at all, and hence they are used delimited in the general case, it is highly recommended to use only characters in names that are valid for identifier names in most programming languages, and that are part of the 7-bit ASCII repertoire, such as [a-zA-Z0-9_], and that the first character of a name isn't a digit; that way, it would be easy to support Muldis D dialects where names are allowed to be non-delimited, as well as more easily permit translation of Muldis D code into other programming languages while making fewer changes. Or if your system is Unicode-saavy (Perl 6 for example), then the recommendation can be broadened to allow the word characters of any script, but names should still avoid including whitespace and punctuation characters.

Type names should all be nouns, or nouns plus adjectives of said, and be named after what a value of that type is; eg, for any given value Foo of type Bar, you can say "Foo is a Bar".

Type names should have all of their words capitalized (first letter of each word is uppercase, other letters of each word are lowercase. It is suggested that the words directly adjoin each other without separator characters, but use of separators is also fine. In other words, see the convention for Perl package names, perhaps aka camel case. But if the type name includes acronyms, then the acronym should be all uppercase, and where necessary, multiple acronyms or abbreviations would then often be underscore-separated. For examples, Int or RatRoundRule.

Function routine names should all be nouns, and be named after the meaning of what they result in; also, function parameters should all be nouns, and be named after the meaning of what they convey to the functions. For example, the integer difference function results in the difference when its subtrahend argument is subtracted from its minuend argument. Every invocation of a function in Muldis D denotes a value, just like any program literal, and it is helpful for it to be named accordingly. A function name should never be a verb, as a function does not take an action or do something or have a side-effect; its invocation is something. Likewise, all named expression names should be nouns.

For a boolean-resulting function, the function name typically is best the text of a question whose answer is just yes or no; for example, is_same or is_a_member. While not a noun per se, such names indicate that the meaning of the function's result is the answer of the question as applied to the function's arguments.

For functions that are best described as deriving Foo from Bar, such as between numbers and strings, they should be named with the result coming first in the name and the input coming second, like Foo_from_Bar; do not use names like Bar_to_Foo. If the convention of prefix/polish notation is also followed, then each part of the function's name is next to what it is describing (output closer to output, input to input, etc), not opposite.

Procedure routine names, by contrast, should all be verbs, and be named after what action they take, because their invocation does do something and/or has a side-effect, and does not denote a value. For example, fail or create_function. Their parameters should be nouns as per function routines. Their variables should also be nouns.

If a routine has just one main parameter, and/or the routine could be conceived as a method for an object that is that argument, and there isn't already a good name for the parameter, then topic is a good name to default to; it says that this argument is the topic that the routine is most concerned with, as per what $_ means to Perl (a topicalizer).

All routine names should have their words separated by underscores, and they generally should be entirely lowercase. The main exception to this is if the routine name embeds a type name, in which case the type should be spelled with its normal casing, such as Int_from_Text.

When a relation is best described as being a plurality of a kind of thing, where each of its tuples is exactly one of those things, then any contexts working with such relations should name the whole relation in the plural form, not the singular form. For example, if each tuple of a relation represents either a person or a business, then a relvar of that relation's type should have a name like "people" or "businesses" rather than "person" or "business"; similarly, the relation type of that relvar might be "People" or "Businesses" while the tuple type which that relation type is defined partly in terms of might be "Person" and "Business".

When defining a pair of closely related types where one of the types is a relation type and the other type is the tuple type over which that relation type is directly partially defined, then: 1. If one of the types is clearly dominant over the other in terms of importance or likelihood to be used as the declared type of anything (most likely the relation type), and especially if it would be a challenge to name the subservient type, then the pair should be grouped together under their own subdepot, such that the subdepot has the conceptual name of the dominant type (and is referenced directly by users as a proxy for said), and the dominant type's actual name under that subdepot's namespace is the empty string, and the subservient type's name under that namespace is either T or R depending on whether it is a tuple or relation type. For example, given the system-defined relation type Array, that type's actual name is Array."" and the tuple type it is partly defined over is named Array.T. 2. If the two types have a mutually even status, and it would be relatively easy to name both of them, then the pair should live out in a more public namespace, each directly under their conceptual names.

Defaults and Options

When defining a routine parameter or type attribute that is expected to have the same single value in a majority of usage scenarios, and that users would generally conceive that they shouldn't have to explicitly specify a value for that parameter or attribute when it would be that common value, then you should give the parameter or attribute an appropriate name and declared type such that the single default value of the declared type is the same as the common value in question; for a parameter, typically it should also be marked as optional.

Similarly, if a certain attribute or parameter would be considered non-applicable or ignorable under certain circumstances, such as defined by neighboring parameters or circumstances, then said ignorables should be named and typed such that their declared type gives them a default value that is best for use when it should be ignored.

One example is a binary choice encoded as a boolean value; the boolean parameter or attribute can be named after the infrequent choice such that giving it a true value will pick the infrequent choice and letting it default to the false value (Bool:False is the default value of the Bool type) will pick the frequent choice.

Generally speaking, the default value of a type should correspond most closely to its concept of empty, unless there is no such concept in the type. So for string types this would mean the empty string, or for collection types, one with no elements, or for numeric types, the value zero.

When declaring a scalar root type that has exactly one possrep, the name of that possrep should be the empty string, which makes it easier for that possrep to be referred to by default in contexts where a possrep name is optional to specify. By contrast, when declaring a scalar root type that has at least 2 possreps of its own, none of the possreps should have the empty string as their name. When declaring a scalar subtype that adds one or more extra possreps, none of the added possreps should have the empty string as its name. On the other hand, don't be hesitant to just use a non-empty name even when there is just one possrep, if that name results in better self-documentation of the type and its value selections.

Syntax Ordering

Muldis D internally (that is, in the system catalog) organizes parts of syntax by name rather than by order, so in that respect the language is not specifically prefix/polish or postfix/reverse-polish or infix or circumfix, and a Muldis D dialect can take any of those forms or mix them. You can even write your code right-to-left if you want, assuming a compatible parser.

That said, the recommended convention is to use prefix/polish notation conceptually, or actually if the dialect makes a distinction. So for example, in a value expression, the result comes out on the left side and the inputs go in on the right side. At least this is assuming you normally read left-to-right. But even if you don't, practically all programming languages are oriented left-to-right anyway, and we can follow that.


Go to Muldis::D for the majority of distribution-internal references, and Muldis::D::SeeAlso for the majority of distribution-external references.


Darren Duncan (


This file is part of the formal specification of the Muldis D language.

Muldis D is Copyright © 2002-2011, Muldis Data Systems, Inc.

See the LICENSE AND COPYRIGHT of Muldis::D for details.


The TRADEMARK POLICY in Muldis::D applies to this file too.


The ACKNOWLEDGEMENTS in Muldis::D apply to this file too.