ASL Lexical Syntax

ARM logo In my post about dissecting the ARM Machine Readable Architecture files I described how to extract the ASL code from the XML files that ARM provides. In this post, I will describe how to start processing that code by examining the lexical syntax. In doing so, I will be going back to one of the first things I had to figure out when I started trying to use ARM’s documentation as an executable specification so I will be looking at code I have barely thought about in 6 years and trying to remember my thought processes at the time as I reverse engineered the language lurking inside ARM’s pseudocode.

There are two aspects to ASL’s lexical syntax: tokens and indentation.


For the most part, ASL is a conventional C-like token syntax. Files are plain ASCII and tokens are defined as follows.

  • Identifiers start with a letter or underscore and continue with zero or more letters, underscores or digits.

    It is almost true that only global variables, constants, functions and user-defined types begin with an uppercase letter and only local variables and formal parameters of functions begin with a lowercase letter, but it is not 100% true. It is also almost true that names don’t begin with an underscore but there are also a few exceptions.

    My implementation also allows dollar-signs in identifiers and it also borrows a trick from Scala and allows any sequence of characters enclosed in backticks to be used as an identifier; this is useful if you want to define builtin functions such as + and /. Neither of these extensions appear in the published specification.

  • Qualified identifiers are identifiers that additionally contain a single decimal point character. For example “AArch32.CallSupervisor”.

    My implementation currently detects these in the parser and treats them as distinct from identifiers because the original plan for how we would use qualified identifiers changed after we started using them. I might go back and clean this up sometime.

  • Integers are written either in decimal using one or more of the characters 0-9 or in hexadecimal using ‘0x’ at the start followed by the characters 0-9, a-f, A-F and underscore. Fixed point numbers are written in decimal and consist of one or more decimal digits, a decimal point and one or more decimal digits.

    The underscores in hexadecimal numbers are not significant and their only purpose is to make large constants such as 0xefff_fffe or 0x8000_0000_0000_0000 easier to read.

  • Constant bit-vectors (called bit-strings in ARM’s documentation) are written using 1, 0 and spaces surrounded by single-quotes. The spaces are not significant and are only used to improve readability of long constants so ‘1111 1111 1111 1111’ is the same as ‘1111111111111111’.

  • Constant bit-masks are written using 1, 0, x and spaces surrounded by single-quotes. The x represents a don’t care character and spaces are only used to improve readability.

  • Strings are surrounded by double-quotes.

    Strings play only a minor role in the architecture specifications so there are no escape mechanisms to allow strings to contain control characters or double quotes.

    In the v7-A specifications, masks were written using both single and double-quotes. In v8-A, masks are only written using single-quotes.

  • The following character sequences are used as delimiters: “(“, “)”, “[”, “]”, “{“, “}”, “+”, “-“, “*”, “/”, “!”, “^”, “&&”, “||”, “==”, “!=”, “<=”, “>=”, “<”, “>”, “<<”, “>>”, “=”, “;”, “,”, “.”, “:”, “..”, “+:”, “&”, “++”.

    Of these, the most unfamiliar is probably “+:” which is used to define a bitslice of a particular width. For example “v<4 +: 8>” represents the 8-bit long bit vector consisting of bits 4, 5, … 11 of variable v. Unlike C, the operator “^” represents exponentiation, not xor.

    A symbol representing implication is in the process of being added - we have not settled on “-->”, “->”, “==>” or “=>”.

  • The following identifiers are reserved words: “if”, “then”, “elsif”, “else”, “case”, “of”, “when”, “otherwise”, “for”, “to”, “downto”, “while”, “do”, “repeat”, “until”, “return”, “assert”, “Consistent”, “enumeration”, “constant”, “is”, “array”, “bit”, “QUOT”, “REM”, “DIV”, “MOD”, “AND”, “NOT”, “OR”, “EOR”, “IN”, “UNDEFINED”, “UNKNOWN”, “UNPREDICTABLE”, “CONSTRAINED_UNPREDICTABLE”, “SEE”, “IMPLEMENTATION_DEFINED”, “SUBARCHITECTURE_DEFINED”, “RAISE”, “try”, “catch”, “throw”, “is”, “typeof”, “TypeOf”.

    “typeof” and “TypeOf” are synonyms — we have not committed to whether to use lowercase or CamelCase but “typeof” is by far the most common.

    “in” is currently implemented as a reserved word as well and is a synonym for “IN” but I believe that we have eliminated all uses of “in” and it could be retired.

    The following identifiers are also implemented but are not found in the published parts of ARM’s specifications. “__type”, “__register”, “__forall”, “__newevent”, “__event”, “__newmap”, “__map”, “__config”, “__intersect”, “__namespace”, “__overloaded”. I will say more about these when describing the ASL grammar.

  • Comments can be written in two styles. The dominant style begins with “//” and lasts until the end of the line but the classic C-style surrounded by “/*” and “*/” is also supported. The latter style can be nested.

In addition to the above, we recognise the “#-line” markers inserted by the C preprocessor to indicate that a file was “#include”d or that a section of code was omitted due to use of “#ifdef”.


ASL follows the example of Occam, Haskell and Python of using indentation to indicate structure instead of requiring some form of “parentheses” such as “{“/”}”, “begin”/”end”, etc.

The way this is implemented is that the we have three additional tokens “NEWLINE”, “INDENT” (indicating an increase in indentation since the last non-empty, non-comment line) and “DEDENT” (indicating a decrease in indentation since the last non-empty, non-comment line).

As we are processing a file, we maintain a stack of indentation depths. On the first token of each line, we compare the indentation of that token with the depth on top of the stack.

  • If the indentation is greater, we emit a single “INDENT” token and push the new indentation onto the stack.
  • If the indentation is less and the stack contains that indentation depth, we pop the stack back to that depth and emit a corresponding number of “DEDENT” tokens.
  • It is an error if the current indentation is not present in the stack.

These indentation rules are inconvenient in multi-line expressions and ASL follows Python by treating multiple lines as a single line inside certain pairs of tokens:

  • ”(“ and “)”; “[” and “]”; “{“ and “}”
  • “if” or “elsif” and “then”
  • “while” and “do”

There is a special circle of hell reserved for those who use tabs in source files. Offenders are doomed to endless arguments with those whose editors use a different tab alignment. That said, the current implementation of ASL follows Haskell and defines tab alignment as 8 but it should really be reported as an error. All ARM specifications use a 4-character indentation increment but that is not required by ASL.

Combining the two

The implementation implements these two parts by having two nested lexers. The inner lexer is a conventional lexer for identifiers, numbers, symbols, reserved words, etc. and tags all tokens with a line number and column number. The outer lexer implements the indentation rules and uses the location information to insert additional NEWLINE, INDENT and DEDENT tokens.

Written on May 7, 2017.
The opinions expressed are my own views and not my employer's.