What does syntax directed translation mean?

Question

Can anyone, in simple terms, explain what does \"Syntax Directed Translation\" mean? I started to read the topic from Dragon Book but couldn\'t understand. The Wiki

Answer 1

In simplest terms, 'Syntax Directed Translation' means driving the entire compilation (translation) process with the syntax recognizer (the parser).

Conceptually, the process of compiling a program (translating it from source code to machine code) starts with a parser that produces a parse tree, and then transforms that parse tree through a sequence of tree or graph transformations, each of which is largely independent, resulting in a final simplified tree or graph that is traversed to produce machine code.

This view, while nice in theory, has a drawback that if you try to implement it directly, enough memory to hold at least two copies of the entire tree or graph is needed. Back when the Dragon Book was written (and when a lot of this theory was hashed out), computer memories were measured in kilobytes, and 64K was a lot. So compiling large programs could be tricky.

With Syntax Directed Translation, you organize all of the graph transformations around the order in which the parser recognizes the parse tree. Instead of producing a complete parse tree, your parser builds little bits of it, and then feeds those bits to the subsequent passes of the compiler, ultimately producing a small piece of machine code, before continuing the parsing process to build the next piece of parse tree. Since only small amounts of the parse tree (or the subsequent graphs) exist at any time, much less memory is required. Since the syntax recognizer is the master sequencer controlling all of this (deciding the order in which things happen), this is called Syntax Directed Translation.

Since this is such an effective way of keeping down memory use, people even redesigned languages to make it easier to do -- the ideal being to have a "Single Pass" compiler that could in fact do the entire process from parsing to machine code generation in a single pass.

Nowadays, memory is not at such a premium, so there's less pressure to force everything into a single pass. Instead you generally use Syntax Direct Translation just for the front end, parsing the syntax, doing typechecking and other semantic checks, and a few simple transformations all from the parser and producing some internal form (three address code, trees, or dags of some kind) and then having separate optimization and back end passes that are independent (and so not syntax directed). Even in this case you might claim that these later passes are at least partly syntax directed, as the compiler may be organized to operate on large pieces of the input (such as entire functions or modules), pushing through all the passes before continuing with the next piece of input.

Tools like yacc are designed around the idea of Syntax Directed Translation -- the tool produces a syntax recognizer that directly runs fragments of code ('actions' in the tool parlance) as productions (fragments of the parse tree) are recognized, without ever creating an actual 'tree'. These actions can directly invoke what are logically later passes in the compiler, and then return to continue parsing. The imperative main loop that drives all of this is the parser's token reading state machine.