Expressions

1.Identifiers

1.1.Regular identifiers

Identifiers are used as names for variables, functions, macros, types and categories. Regular identifiers should match the regular expression [a-zA-Z_]+[a-zA-Z0-9_$?]*. That is, identifiers should

In addition, it is customary to use the following guidelines when choosing names:

Besides the regular identifiers, Mathemagix allows the programmer to use several types of special identifiers for the names of operators and special objects.

1.2.Operators

First of all, identifiers corresponding to the built-in operators (see section ? below) are formed by prefixing them by one of the keywords prefix, postfix, infix and operator. For instance:

postfix ! (n: Int): Int == if n=0 then 1 else n * (n-1)!;

Similarly, the instruction

mmout << map (infix *, [ 1, 2, 3 ], [ 4, 5, 6 ]) << lf;

prints the vector . The operator

operator []: (t: Tuple T) -> Vector T;

is used as a shorthand for the constructor of vectors (so that we may write [ 1, 2, 3 ] instead of vector (1, 2, 3)). Similarly, the operator postfix [] is used for accessing entries of vectors and matrices.

Remark 1. TODO: in place operators formed with the keyword inplace.

1.3.Named access operators

In addition to the builtin operators, any regular identifier id can also be turned into a postfix operator postfix .id. For instance, when defining a class Point by

class Point == {
  x: Double;
  y: Double;
  constructor point (x2: Double, y2: Double) == {
    x == x2;
    y == y2;
  }
}

Mathemagix automatically creates two such postfix operators for accessing the fields x and y:

postfix .x: Point -> Double;
postfix .y: Point -> Double;

Hence, we may define an addition on points using

infix + (p: Point, q: Point): Point ==
  point (p.x + q.x, p.y + q.y);

Additional operators similar to postfix .x and postfix .y can be defined outside the class

postfix .length (p: Point): Double ==
  sqrt (square p.x + square p.y);

Given a point p, we then write p.length for its length as a vector.

1.4.Other identifiers

The Mathemagix keywords, such as while, class, etc. can also be turned into identifiers by prefixing them with keyword. Hence, keyword while stands for the keyword while. This notation is mainly using during formal manipulations of Mathemagix programs.

More generally, any valid string can be turned into an identifier by putting it between quotes and prefixing it by the keyword literal. For instance, literal "sqrt" is equivalent to the regular identifier sqrt, literal "+" is equivalent to the infix operator infix +, and literal "_+_" is an identifier which can only be written using the keyword literal.

We finally notice that this is a special identifier which denotes the underlying instance inside a class method.

2.Literal constants

Mathemagix provides three types of literal constants: string literals, integer literals and floating literals. In addition, there are several important constants, such as true and false, which are really identifiers from the syntactic point of view.

2.1.Literal strings

Short string constants are either written inside a pair of double quotes "...":

mmout << "Hello world" << lf;

Double quotes and backslashes inside strings should be escaped using backslashes:

quote_char    : String == "\"";
backslash_char: String == "\\";

Long string constants which avoid this kind of escaping can be formed using the delimiters /"..."/, as in the following example:

hello_world_example: String == /"
include "basix/fundamental.mmx";
mmout << "Hello world!" << lf;
"/
mmout << hello_world_example << lf;

2.2.Literal integers

An integer literal is a sequence of digits, possible preceded by a minus sign. It matches the regular expression [-]?[0-9]+. Examples are: 123456789123456789, -123. The user should define a routine

literal_integer: Literal -> T;

in order to allow literal integers to be interpreted as constants of type T. The file basix/int.mmx of the standard library defines the routine

literal_integer: Literal -> Int;

which allows literal integers to be interpreted as machine integer constants. Arbitrary precision integers are supported by importing

literal_integer: Literal -> Integer;

for numerix/integer.mmx.

2.3.Literal floating point numbers

A literal floating point constant is a sequence of digits with a decimal point inside, an optional sign and an optionalexponent. It matches the regular expression

[-]?[0-9]+[.][0-9]+[[eE][-]?[0-9]+]?

The user should define a routine

literal_floating: Literal -> T;

in order to allow literal floating poiunt numbers to be interpreted as constants of type T. In particular, the files basix/double.mmx and numerix/floating.mmx from the standard library define the routines

literal_floating: Literal -> Double;    // in basix/double.mmx
literal_floating: Literal -> Floating;  // in numerix/floating.mmx

For instance,

zero : Double   == 0.0;
pi   : Double   == -3.14159;
funny: Floating == 1.2345679012345678901234567890e2012;

Notice that 0. is not permited: one must write 0.0.

2.4.Special constants

Some constants are encountered so frequently, that it is useful to mention them here, even though they are really identifiers from the syntactic point of view:

3.Operators

Table ? summarizes all standard Mathemagix operators, together with their binding forces. For instance, the expression

a*b + c > d

is naturally parsed as . The operators can roughly be divided into four groups:

  1. Infix operators such as + apply to one argument on the left and one argument on the right.

  2. Prefix operators such as negation prefix ! apply to one argument on the right.

  3. Postfix operators such as the factorial postfix ! apply to one argument on the left.

  4. Other special operators, such as operator [] for writing vectors [ 1, 2, 3 ].

There are also some special postfix operators, such as function application postfix () and named access operators such as postfix .x (see section ?). Most operators are infix operators, so infix notation is assumed to be the default, unless stated otherwise. In the remainder of this section, we will quickly survey the intended purpose of the various operators.

Assignment operators ==, :=, +=, -=, *=, /=, <<=, >>=, ==>, :=>
Function expressions lambda, :->
Input/output operators <<, >>, <<<, >>>, <<*, <<%
Logical implication =>, <=>
Logical disjunction or, \/, xor
Logical conjunction and, /\
Relations =, <, >, <=, >=, !=, !<, !>, !<=, !>=, :, in
Type conversion :>, ::, ::>
Arrows ->, ~>
Ranges .., to, downto
Addition and subtraction +, -, @+, @-
Multiplication and division *, /, @*, @/, div, quo, rem, mod, @, ><, %, &
Prefix operators !, ++, , -, @-, @, #, &
Operate on empty string
Power ^
Postfix operators ++, , !, ', , ~, #, (), []
Tuples and vectors (), []

Table 1. Overview of all Mathemagix operators listed by increasing binding force.

3.1.Assignment operators

The operators == and := are used for declarations of constants and mutable variables, as described in the sections about the declaration of variables and functions. The operator ==> is used for macro definitions (see the section about macro declarations). The operator :=> is reserved for future use.

The operator := can always be used for the assignment of mutable variables. The operators +=, -=, *=, /=, <<= and >>= are not yet exploited in the standard libraries, but there intended use is “assignment of the left hand expression with the result of the corresponding outplace operator applied to both arguments”. For instance, the instruction

a += b;

should considered to be equivalent to the assignment

a := a + b;

Remark 2. As a future extension of the compiler, we also intend to support assignment to tuples, in order to assign several mutable variables at once. For instance,

(a, b) := (b, a)

should swap the variables a and b, and

(a, b) += (x, y)

should respectively increase a and b with x and y.

3.2.Function expressions

The special operators :-> and lambda are used for writing functions directly as expressions. The expressions

(a_1: T_1, ..., a_n: T_n) :-> (val: Ret_Type)
lambda (a_1: T_1, ..., a_n: T_n): Ret_Type do val

can both be used as a notation for the function with arguments a_1, , a_n of types T_1, , T_n, which returns the value val of type Ret_Type.

3.3.Input/output operators

The operators << and >> are respectively used for sending data to an output port and retrieving data from an input port. The same notation is useful in analogous circumstances, such as appending data to a vector or popping data from a stack.

The operators <<< and >>> are used for sending and retrieving data in binary form. This allows for instance for the implementation of efficient communication protocols between different processes on the same or distant computers. The operators <<* and <<% are reserved for future use.

Remark 3. Sometimes, we also use the operators << and >> as shift operators. For instance, given a power series f in , we might write f << n as a shorthand for the multiplication of f with . However, it should be noticed that the binding force of << and >> is not really appropriate for this type of use (a binding force similar to the one of multiplication would be better for this kind of use), so one carefully has to put brackets whereever necessary in this case. In future versions of Mathemagix, this kind of overuse of notations might be discouraged.

3.4.Logical operators

The operators =>, <=>, \/, /\ stand for the standard logical connectors , , and . The prefix operator prefix ! stands for logical negation . These operators are functions which can be redefined by the user, so both arguments are evaluated in case of the logical connectors.

Mathemagix also provides the built-in operators or and and which must take arguments of type Boolean. Moreover the second argument of or is evaluated only if the first argument evaluates to false. Similarly, the second argument of and is evaluated only if the first argument evaluates to true.

Remark 4. The operators =>, <=>, \/, /\ and ! can also be useful for bitwise operations on numbers, even though the binding force is someone inappropriate for this kind of use. One might also want to use /\ as a notation for exterior products, again with an inappropriate binding force. In future versions of Mathemagix, this kind of overuse of notations might be discouraged.

3.5.Relations

The operators =, < , >, <= and >= correspond to the standard mathematical relations , , , and . When prefixing these relations with !, we obtain the operators !=, !<, !>, !<=, !>= which correspond to their negations , , , and .

In computational contexts, mathematical relations often admit a very asymmetric nature: typically, it can be very easy to prove inequality, but very hard to prove equality. It can even happen that we have an algorithm for proving inequality, but no known algorithm for proving equality. This is for instance the case for the class of so called exp-log constants, constructed from the rational numbers using the field operations, exponentiation and logarithm. In contexts where equality testing is hard, it is therefore useful to make a notational distinction between various types of equality, such as proven equality, probable equality, syntactic equality, etc.

In Mathemagix, the intention of the notations =, <, >, <= and >= is that they stand for proven relations. On the other hand, the negations !=, !<, !>, !<= and !>= are intended to be mere shortcuts for their (not necessarily proven) negations. Hence, a != b should always be equivalent to !(a = b). We are working on a comprehensive set of additional relations for proven negations; they will probably be denoted by =/, </, >/, <=/, >=/. As an additional rule, it is our intention that <= (resp. >=) is always equivalent to the disjunction of < (resp. >) and =. Thus a <= b should always be equivalent to the statement a < b or a = b.

The operator : should be read “is of type”. For instance, x: T stands for “x is of type T”. The operator in occurs inside the for-in construct (see the section about loops).

3.6.Type conversion

The operator :> can be used to convert an expression of a given type into another, provided that an appropriate converter was defined. More precisely, assume that expr has type S and that we defined a converter convert: S -> D. Then expr :> D stands for the explicit conversion of expr into an expression of type D. More information about type conversions can be found in the section on explicit type conversions and user defined converters.

3.7.Arrows

The operator -> is used as an efficient notation for function types, such as Int -> Int. One typical use case of this notation is when a function is passed as an argument to another function:

iterate (f: Int -> Int, n: Int) (k: Int) ==
  if n = 0 then k else iterate (f, n-1) (f k);

The operator ~> is mainly used as a notation for key-value bindings whenever we explicitly wish to create a table with given entries. For instance:

basic_colors: Table (String, Color) ==
  table ("red" ~> rgb (1, 0, 0),
         "green" ~> rgb (0, 1, 0),
         "blue" ~> rgb (0, 0, 1),
         "white" ~> rgb (1, 1, 1));

or

forall (T: Type)
invert (t: Table (T, T)): Table (T, T) ==
  table (t[key] ~> key | key: T in t);

3.8.Ranges

There are three standard kinds of range operators:

start to end Range from start up to end included
start .. end Range from start up to end not included
start downto end Range from start down to end included

3.9.Arithmetic operations

The standard arithmetic operations +, -, *, / and ^ stand for addition, subtraction, multiplication, division and powering. The @-prefixed variants @+, @-, @*, @/ stand for , , and . Notice that - and @- can either be infix or prefix operators.

Mathemagix provides the additional operators div, quo, rem and mod for division-related operations in rings which are not necessarily fields. The operator div stands for (usually partially defined) exact division. For instance, numerix/integer.mmx provides the operation

infix div: (Integer, Integer): Integer;

but 5 div 3 is undefined and might raise an error. The operators quo and rem stand for quotient and remainder in euclidean domains. Hence, we should always have

The operator mod stands for modular reduction, so that the return type is usually different from the source types. For instance 5 mod 3 would typically belong to Modular (Int, 3) or Modular (Integer, 3).

There are a few other operations with the same binding force as multiplication. The append operator ><, also denoted by , is typically used for appending strings, vectors and table. For instance "a" >< "b" yields "ab". The operator infix @ is used for functional composition, whereas the operators % and & are reserved for future use.

3.10.Prefix operators

The standard prefix operators in Mathemagix are prefix ! (negation ), prefix ++ (increment), prefix – (decrement), prefix - (unary ), prefix @- (unary ), prefix @ (explode), prefix # (size), prefix & (reserved for future use).

In addition, operator application of the form sin x parses in a similar way as when sin behaves as a prefix operator. For instance, sin cos x should be parsed as .

3.11.Postfix operators

The standard postfix operators in Mathemagix are postfix ++ (post increment), postfix – (post decrement), postfix ! (factorial ), postfix ' (quote or derivative), postfix ‘ (unquote), postfix ~ and postfix # (reserved for future use).

++, , !, ', , ~, #, (), []

In addition, Mathemagix provides the special postfix operators postfix () and postfix [] for which we are allowed to put additional arguments between the brackets. Hence, postfix () stands for the traditional notation of function application, whereas postfix [] is typically used as an accessor for compound data structures. Notice that f(g)(x) is parsed as , whereas f g x is parsed as .

Using the operator postfix (), we may generalize the classical notation for function application to user defined types, such as vectors of functions:

postfix () (v: Vector (Int -> Int), x: Int): Vector Int ==
  [ f x | f: Int -> Int in v ];

3.12.Tuples and vectors

The reserved special operator operator () is used for building tuples of expressions (of possibly different types), such as (1, "hello"). The special operator operator [] is used as a notation for explicit vectors, such as [ 1, 2, 3 ], but it might be used for other purposes.

4.Generators

Generators are an elegant way for representing a stream of data of the same type. For instance, the expression 1 to 10 of type Generator Int allows us to write

for i: Int in 1 to 10 do
  mmout << i << " * " << 7 << " = " << 7*i << lf;

Mathemagix provides several constructs for forming generators.

4.1.Range generators

There are three standard kinds of range operators:

start to end Range from start up to end included
start .. end Range from start up to end not included
start downto end Range from start down to end included

4.2.The explode operator

Many container types come with a prefix operator @ which returns a generator. For instance, given a vector v of type Vector T, the expression @v has type Generator T. Whenever expr is an expression such that @expr has type Generator T, we are still allowed to use expr as the in-part of the for-in construct. For instance:

for i: Int in [ 2, 3, 5, 7, 11, 13, 19 ] do
  mmout << i*i << lf;

4.3.The such that construct

One other important construct for forming generators is the “such that” operator |. Given a generator g of type Generator T, the expression

(var: T in g | predicate? var)

stands for the generator of all items in g which satisfy the predicate predicate?. For instance, consider the following naive implementation of the predicate prime? which checks whether a number is prime

prime? (n: Int): Boolean == {
  for i: Int in 2..n do
    if i rem i = 0 then return false;
  return true;
}

Then we may display the vector of all prime numbers below using

mmout << [ p: Int in 1 to 1000 | prime? p ] << lf;

Notice that this vector is constructed from the expression

(p: Int in 1 to 1000 | prime? p)

of type Generator Int using the bracket operator operator [].

4.4.The where construct

The vertical bar | can also be used as the “where” operator, using the following syntax:

(expr var | var: T in g, predicate_1? var, ..., predicate_n? var)

Here g is again a generator of type Generator T, expr var any expressions which involves var, and predicate_1? var, ..., predicate_n? var an arbitrary number of predicates which involve var. If expr var has type U, then the resulting expression has type Generator U. For instance,

mmout << [ i^2 | i: in 1 to 100 ] << lf;

displays the vector of all squares of numbers from to , and

mmout << [ i^2 | i: in 1 to 1000, prime? (4*i + 3) ]

displays the square of each number such that is prime.

4.5.The where construct with multiple generators

Mathemagix actually supports a generalization of the where construct with multiple generators and predicates at the right-hand side. This generalization is best illustrated with an example:

mmout << [ p^i | p: Int in 1 to 1000, prime? p,
                 i: Int in 1 to 10, p^i < 1000 ] << lf;

This code will print the unordered vector of all prime powers below .

4.6.Matrix notation

A special where notation || is used for generators which allow to build rows of matrices or similar two dimensional structures. Again, this notation is best illustrated with an example. Assuming that the file algebramix/matrix.mmx was included, the expression

[i+j | i: Int in 0 to 9 || j: Int in 0..10]

computes the following matrix:

5.Mappers

Most Mathemagix containers implement a mapping construct map. This construct is used for applying one or more functions to all entries of one or more containers.

Two simple examples for containers with a single parameter are

mmout << map (square, [ 1, 2; 3, 4 ]) << lf;
mmout << map (infix *, [ 1, 2, 3], [4, 5, 6]) << lf;

These instruction respectively output the matrix and the vector .

In the case of containers with more than one type parameter, one usually has to provide one mapping function for every such parameter. Consider for instance the following table:

t: Table (Int, String) == table (3 ~> "Hello", 4 ~> "Hi", 8 ~> "Bonjour");

Then the instruction

mmout << map (square, prefix #, t) << lf;

prints the table .

Syntactically speaking, the construct map is an ordinary identifier. For instance, assuming that we defined a container Complex R (see the section on how to define your own containers), a unary mapper for this container can be defined as follows:

forall (R1: Ring, R2: Ring)
map (f: R1 -> R2, z: Complex R1): Complex R2 ==
  complex (f z.re, f z.im);

When providing your own containers, it is actually important to define unary mappers of this kind, because such mappers automatically induce converters between containers of the same kind but with different type parameters. For instance, given a converter from R1 to R2, the above mapper for complex numbers automatically induces a converter from Complex R1 to Complex R2. This allows the user to write

z: Complex Rational == complex (1, 2);

In general, such a converter is constructed whenever the user provides a unary mapper which takes one mapping function for each parameter on input together with a single container.

Remark 5. We notice that the existence of a unary mapper is mandatory if a program both uses the container in an generic and in a specialized way, and if conversions between the generic and specialized versions of the container indeed occur in the program. For instance, some mathematical library lib.mmx might provide a generic function

forall (R: Ring)
conj (z: Complex R): Complex R == complex (z.re, -z.im);

Now assume that we a client program client.mmx which only works with complex numbers of type Complex Double and which has specialized this type for better performance. In memory, this means that such complex numbers are represented by pairs of double precision numbers rather than pairs of pointers to double precision numbers numbers as would be the case for generic complex numbers. However, the routine conj from lib.mmx a priori only applies to generic complex numbers, so conversions between the specialized and the generic view are necessary if we want to use this routine in client.mmx. As soon as the required unary mapper is defined, these conversions are automatic.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU General Public License. If you don't have this file, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.