Questions: Compiler Phases and Organization

Question 1

A compiler reports 'variable x used before declaration.' Which phase produces this error?

• A. Lexical analysis
• B. Syntax analysis (parsing)
• C. Semantic analysis
• D. Code generation

Answer: (C)

Lexical analysis only identifies tokens — it cannot know what a name means or whether it was declared. Syntax analysis checks grammatical structure but has no concept of declarations or scope. Semantic analysis is responsible for name resolution, scope checking, and type checking; it uses the symbol table to track declared names and detect uses of undeclared or out-of-scope variables.

Question 2

A one-pass compiler that processes source code from top to bottom exactly once is architecturally impractical for modern programming languages.

• T. True
• F. False

Answer: False

One-pass compilers have been built and work correctly — Pascal was famously compiled in one pass. However, they require the language to be designed so that definitions always precede uses (no forward references). Modern languages like C and Java allow calling a function before its definition appears in the file, which forces multi-pass compilation. The statement is false because 'impossible' is too strong: it is a language design choice, not a physical constraint.

Question 3

Why is an intermediate representation (IR) phase valuable even when a compiler targets only a single machine architecture?

Model answer: IR decouples the front end (language-specific parsing and semantic analysis) from the back end (machine-specific code generation). Even for a single target, IR enables optimization passes expressed at a level above machine code but below source code — constant folding, dead code elimination, inlining, and loop transformations are all easier to implement on a clean IR than on raw assembly. The IR is also the right level for analyses that require seeing the whole program at once.

The IR is the architectural boundary that makes the classic front end / middle end / back end split work. Optimizations written against the IR remain correct regardless of the source language or target machine. Without IR, every optimization would need to be reimplemented for each source-target combination. Even for single-target compilers, the IR pays for itself by making the optimization infrastructure reusable and testable independently of code generation.