[ondrik]

Summary

Addresses most of #57 (the remaining item is listed below).

A cleanup of the repository with no change in behaviour: stray and obsolete files, and the items of #57 that are pure cleanups. Each commit covers one topic, so the PR is easiest to review commit by commit.

Files

• Remove .DS_Store (now ignored), .travis.yml (Travis configuration for GCC 4.8; CI runs on GitHub Actions) and .vscode/settings.json.
• Remove unit_tests/reference_answers/. No test or script uses it. Its .aut files use a header the current parser rejects, and artifact_evaluation.sh calls a command-line syntax that no longer exists.
• .gitignore: drop entries for directories that no longer exist (python_interface, backup, equivalence_or_not, dyadic, revlib, POPL24).

Code (#57)

Dead code

• Remove AUTOQ::Constraint and AUTOQ::check_validity, which have no callers.
• Remove the CLI's SIGALRM timeout handler and its globals; it was never armed. Remove the unused Aut alias.
• Remove -DCACHE_AC -DCACHE_OPT_CONGR from cli/CMakeLists.txt; no source file reads them.
• Remove the no-op || find("Predicates") clause in ReadTwoAutomata.
• Delete the old hand-written parser headers (predicate_parser.hh, complex_parser.hh, constraint_parser.hh, symboliccomplex_parser.hh) and qualify the three uses of Complex that relied on their global using.
• Remove switch_symbol_to_second and the unused second symbol type of EvaluationVisitor/TimbukParser: TimbukParser<Symbol, Symbol> is now TimbukParser<Symbol>. Also remove the commented-out childStateMap code in CX, CZ and CCX.

Duplicated logic

• The inclusion search shared by Concrete and Symbolic automata now lives in include/autoq/inclusion_bfs.hh. Each variant supplies a small leaf policy:

  • Concrete checks that B's amplitudes are a common nonzero multiple of A's;
  • Symbolic collects a vertex's leaf pairs and asks the SMT solver.

  The inclusion-time statistics are recorded as before. The Index variant enumerates B's transitions differently and stays separate; only its unused possible_colors_for_qA map is removed.

• Keep one duration formatter, Convert::ToString. The only visible difference is in the per-gate debug log: a gate taking between 1 h and 1 day is logged as 1h5m5s instead of 0d1h5m5s.

• Merge the two identical branches of ConstrainedComplex::operator* and of the unit-test helper run_single_folder_benchmark.

• Move the Extended Dirac header scan and the constraints-to-SMT loop, which were each written twice, into helpers.

Small inefficiencies

• The ratio check in the concrete inclusion ran four times per leaf pair; it now runs once.
• mapped_vector comparisons no longer copy both maps.
• run_verification reuses the parsed precondition instead of parsing pre.hsl again.

Not in this PR

• Merging the four tree-duplication loops (CX with c < t, CZ, CCX with c < c2 < t, diagonal_gate). They differ in details that change the resulting automata:

  • diagonal_gate creates an empty transition entry for every copied state;
  • the loop bounds differ by the hasLoop level;
  • CX swaps children at the target, while diagonal_gate rescales leaves;
  • CZ and CCX keep three copies.

  Merging them is a rewrite of the gate algorithms and deserves its own PR.

• run_execution parsing pre twice through ReadTwoAutomata(pre, pre, circuit). That call also computes the joint qubit order, so it is not a pure cleanup.

• docs/old_README.md (the AutoQ 1.0 README) is kept. It may still serve as the CAV'23 artifact's documentation.

Testing

• Native Release build: compiles, and ctest passes (2/2).
• Output comparison against main: I ran 648 benchmark instances with both ver and ex. They are drawn from every family in benchmarks/ and unit_tests/testcase: all instances of the smaller families and of the per-paper folders, and for the 1000-instance families the first 15 and every 50th. Each run had a 20 s limit.
  • 645 runs finished on main, and all 645 outputs are identical with this branch, ignoring time and memory figures:
    • 498 ver verdicts: 462 [OK], 36 [failed];
    • 147 ex runs, each printing the full resulting set of states.
  • The other 645 runs hit the time limit and were not compared.

🤖 Generated with Claude Code

https://claude.ai/code/session_013gtFQRPe1eSSpmA7EJ2KNY