The assembly process begins with yasm calling the do_parse routine implemented by the parser module. The primary parameters given do_parse include the object to parse into, the preprocessor to use, and the initial input file and filename. All of the output of the parser goes into the object; this consists primarily of a list of sections.

To build the sections and their constituent bytecodes, the parser must call functions provided by the architecture module (to identify instructions and registers) and the object format module (to create sections, implement the global, extern, common directives as well as any other object format specific directives such as ORG). The parser is expected to obtain its text input from the preprocessor moudle.

Usually the parser is implemented in two distinct portions: a tokenizer that breaks the inputs into discrete chunks (tokens) such as identifiers, separators (such as comma and semicolon), and numbers, and a parser that looks for certain sequences of tokens to generate the output. See a compiler book such as [AhoSethiUllman96] for more details.

All parsers need to use the parse_check_id function provided by the architecture module to determine if a particular text identifier is an instruction or register. Some assembler syntaxes may be able to infer this by usage, but it is still necessary for them to call parse_check_id in order to validate this assumption and to obtain the information required by the architecture to later recognize the instruction or register. The architecture is required to pass all information needed to uniquely identify an instruction or register through the 4-word data parameter passed to parse_check_id. The parser in return is expected to save this information in the standard insn bytecode or its operands.

In order to separate the parser module from in-depth parsing of an instruction and its operands, the parser is expected to form insn bytecodes that contain the arch data for the parsed instruction (from parse_check_id) and a list of yasm_insn_operand structures. Each operand may be designated as a register (in which case the arch data from the register’s parse_check_id call is required), an immediate value or expression (as a yasm_expr), or a memory location (as a yasm_effaddr).