Reserved Sequences (Keywords)

The following character sequences are considered reserved, and should be yield a distinct token (rather than be counted as an identifier).

and

bool

custom

else

eh?

false

fromconsole

if

immutable

int

or

otherwise

means

toconsole

return

true

void

while

Identifiers

Any sequence of one or more letters and/or digits, and/or underscores, starting with a letter or underscore, should be treated as an identifier token. Identifiers must not be reserved sequences, but may include a proper substring that is a reserved word, e.g. while1 is an identifier but while is not.

Integer Literals

Any sequence of of one or more digits yields an integer literal token as long as it is not part of an identifer or string.

String Literals

Any string literal (a sequence of zero or more string characters surrounded by double quotes) should yield a string literal token. A string character is either

• an escaped character: a backslash followed by any one of the following characters:
  1. n
  2. t
  3. a double quote
  4. another backslash
  or
• a single character other than newline or double quote or backslash.

Examples of legal string literals:

""
"&!88"
"use \n to denote a newline character"
"use \" to  for a quote and \\ for a backslash"

Examples of things that are not legal string literals:

"unterminated
"also unterminated \"
"backslash followed by space: \ is not allowed"
"bad escaped character: \a AND not terminated

Symbol Operators

Any of the following one- or two-character symbols constitute a distinct token:

=    :    ,    +    -    
==    >    >=    {    <    <=    
(    !    &    !=    --    ++    
}    )    ;    /    *    ->    

Comments

Text starting with # up to the end of the line is a comment (except of course if those characters are inside a string literal). For example:

 
# this is a comment
# and so is # this
# and so is # this %$!#

The scanner should recognize and ignore comments (there is no COMMENT token).

Whitespace

Spaces, tabs, and newline characters are whitespace. Whitespace separates tokens and changes the character counter, but should otherwise be ignored (except inside a string literal).

Illegal Characters

Any character that is not whitespace and is not part of a token or comment is illegal.

Length Limits

No limit may be assumed on the lengths of identifiers, string literals, integer literals, comments, etc. other than those limits imposed by the underlying implementation of the compiler's host language.

For the most part, the token to produce should be self-explanatory. For example, the + symbol should produce the CROSS token, the -  symbol should produce the DASH token, etc. The set of tokens can be found in frontend.hh or in the switch in tokens.cpp. The LCURLY token refers to a left curly brace, {. the RCURLY refers to a right curly brace, }.

The lexical structure of A is very similar to C, with a couple of small alterations:

• The string -> produces the ARROW token.
• The sequences custom, means, otherwise, and eh?, are keywords of the language. They produce the CUSTOM, MEANS, OTHERWISE, and EH, tokens, respectively.
• The string && and || are NOT in the language . Instead "logical and" is represented by the string and and "logical or" is represented by the string or.

Notable Differences from C

While the canonical reference for A syntax is its context-free grammar, there are a couple of "standout" points which deserve special attention for their deviation from C:

i:int;
i = 4;

i:int;
fn : () -> void {
  i = 4;
}

Any type is promotable to itself.

The following shows the atomic operands and the types that they are assigned:

• numeric literals (e.g. 7, 3) are of type int
• bool literals (i.e. true, false) are of type bool
• string literals are of string type
• identifiers have the type of their declaration, which is determined during name analysis

The operators in the language are divided into the following categories:

• logical: not, and, or
• arithmetic: plus, minus, times, divide, negation, postincrement, postdecrement
• equality: equals, not equals
• relational: less than (<), greater than (>), less then or equals (<=), greater than or equals (>=)
• field access: dereference (--, when not used for postdecrement)
• assignment: assign (=)

The type rules of the language are as follows:

• logical operators and conditions are legal if and only if:

  • All operands are bool

  The result type is bool in legal cases, ERROR otherwise.

• arithmetic operations are legal if and only if:
  • Operands are both int - the result type is int

  In all illegal cases, the result type is ERROR.

• relational operations are legal if and only if:

  • Both operands are int

  The result type is bool in legal cases, ERROR otherwise.

• equality operations are legal if and only if:

  • Both operands are of the same primitive type
  • Neither operands are of the same class type
  and
  • Neither operand is a function type

  The result type is bool in legal cases, ERROR otherwise.

• assignment operations are legal if and only if:

  • Both types are the same and the LHS is an lvalue - the result type is that of the LHS

  It is ILLEGAL for the operand of an assignment to be a class name or instance of a class.

  The result type is that of the LHS operand in legal cases, ERROR otherwise.

• consolein operations are legal if and only if:

  • The operand is of int or bool . .

• consoleout operations are legal if and only if:

  • The operand is of type int
  • The operand is of type bool
  • The operand is of type string

• member access operations are legal if and only if:

  • The index operand is a field name of the base class type

  The result type is the type of the field in legal cases, and ERROR otherwise.

• function calls are legal if and only if:

  • The identifier is of function type (i.e., the callee is actually a function).
  and
  • The number of actuals must match the number of formals.
  and
  • The type of each actual must match the type of the corresponding formal.

  If the identifier is not a function name, the result type is ERROR. Otherwise, it is the return type of the function even if the arguments are ill-typed.

• function returns: The follow rules hold for function returns:

  • A return statement in a void function may not have a value.
  • A return statement in a non-void function must have a value.
  • The return expression must match the function's declared type.

  It is LEGAL for a non-void function to skip a return statement. For example, this code is ok:

  f : () -> int {
    //I should return an int, but I don't
}