Downloads: Document (PDF) · Document (DOCX) · Slides (PDF)

AuxFlow: Review 2 Viva Preparation

Static Analysis for Multi-Pass Document Compilation · Upayan Mazumder (24BDS0367) · Compiler Design Lab

Companion guide for Review 2: Front-End, Semantic Analysis, and Intermediate Representation (Phases 1 to 7). All 16 implementation phases are complete. Tests: 245 pass, 1 skipped. mypy –strict is clean on 41 files. ruff lint is clean on src, tests, and bench. This review presents Phases 1 to 7. Formal assessment document: docs/review2-document.docx and docs/review2-document.pdf. Framing remains conservative: AuxFlow is a compiler-based static analysis of dependency structure in multi-pass compilation.

1. What does the lexer do differently from standard compilers?

  • Implements the 16 TeX category codes (catcodes) via a table-driven scanner in auxflow.lexer.
  • Catcodes define lexical interpretation dynamically per character class.
  • Recognizes comments (% vs \%) and inline verbatim blocks (\verb|...|) without expansion.
  • Verifies full round-trip preservation from source characters to tokens.

2. How does the parser handle LaTeX without a full TeX engine?

  • Hand-written recursive descent with one token of lookahead (auxflow.parser.recursive).
  • Operates strictly over a declared language subset; unsupported tokens map to Opaque nodes.
  • Uses an argument signature table (auxflow.parser.signatures) to assign mandatory and optional groups.
  • Employs fault-tolerant error recovery: malformed input never raises exceptions and always produces an AST.

3. How are multi-file documents structured and analyzed?

  • The parser builds an AST per file and records file inclusions (\input and \include).
  • auxflow.ast.forest links file ASTs into a multi-file dependency forest.
  • DFS coloring detects recursive inclusion cycles and isolates unreachable files.

4. What symbol tables exist, and why use two-pass resolution?

  • Six symbol tables: LabelTable, RefTable, CiteTable, PackageTable, FileTable, CounterTable.
  • First pass (declare): records all definitions across all files without binding uses.
  • Second pass (resolve): binds \ref, \pageref, and \cite uses across file boundaries.
  • Two-pass traversal prevents forward references across files from reporting as undefined.

5. What semantic checks run before compilation?

  • auxflow check verifies semantic consistency with zero engine executions.
  • Detects undefined references, missing citations, duplicate labels, unresolvable file includes, and missing graphics.
  • Warns on unmodelled packages via PackageTable before building the graph.

6. What is the Auxiliary Reference Graph (ARG)?

  • The compiler’s Intermediate Representation (IR), implemented in auxflow.ir.
  • Nodes represent instance-level auxiliary facts (LABEL_NUMBER, LABEL_PAGE, PAGEREF_USE, TOC_ENTRY, BBL).
  • Edges represent lagged data dependencies (AUX_ROUNDTRIP, TOOL, LAYOUT, STRUCTURAL) with pass lags.
  • Every node and edge carries a mandatory source span for diagnostic traceability.

7. What is the exists / value split in the IR?

  • List entries (\contentsline) split into two distinct facts:
    1. exists: presence of the entry, determined by structure, feeding CONTENT_VOLUME.
    2. value: typeset page number of the entry.
  • Decouples pagination shift from reference consumption, preventing ordinary tables of contents from forming false cycles.

8. How can we inspect the compiler pipeline at Review 2?

  • Lexer: auxflow tokens FILE
  • AST: auxflow ast FILE
  • Symbol tables: auxflow symbols FILE
  • Semantic analysis: auxflow check FILE
  • Intermediate representation: auxflow graph FILE (exports Graphviz DOT format)

Quick Backups (Viva Questions)

  • Why not use Flex/Bison? TeX catcodes make grammar context-sensitive; hand-written lookahead handles delimiter-driven verbatim and macro arguments reliably.
  • How are spans preserved? Every AST node is a frozen dataclass with a mandatory Span(file_id, start, end).
  • Does semantic check execute TeX? No. It performs pure static symbol table lookup in linear time.
  • What is the complexity of IR lowering? $O(V + E)$ single-pass traversal over the resolved AST forest.