Reduce regular graph construction memory and support NFAs

.github/workflows/build.yml

@@ -428,9 +428,9 @@ jobs:
       uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4
       with:
         path: ${{ runner.temp }}\vcpkg-binary-cache
-        key: vcpkg-${{ runner.os }}-x64-windows-d015e31e90838a4c9dfa3eed45979bc70d9357fc-${{ hashFiles('vcpkg.json') }}
+        key: vcpkg-${{ runner.os }}-x64-windows-45f9f39362a4c52e2b1fbe57b7e649db7f3d96d4-${{ hashFiles('vcpkg.json') }}
         restore-keys: |
-          vcpkg-${{ runner.os }}-x64-windows-d015e31e90838a4c9dfa3eed45979bc70d9357fc-
+          vcpkg-${{ runner.os }}-x64-windows-45f9f39362a4c52e2b1fbe57b7e649db7f3d96d4-
 
     - name: Setup vcpkg
       shell: pwsh
@@ -439,7 +439,7 @@ jobs:
         $vcpkgBinaryCache = Join-Path $env:RUNNER_TEMP "vcpkg-binary-cache"
         New-Item -ItemType Directory -Force -Path $vcpkgBinaryCache | Out-Null
         git clone https://github.com/microsoft/vcpkg $vcpkgRoot
-        git -C $vcpkgRoot checkout d015e31e90838a4c9dfa3eed45979bc70d9357fc
+        git -C $vcpkgRoot checkout 45f9f39362a4c52e2b1fbe57b7e649db7f3d96d4
         & (Join-Path $vcpkgRoot "bootstrap-vcpkg.bat") -disableMetrics
         "VCPKG_ROOT=$vcpkgRoot" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append
         "VCPKG_DEFAULT_BINARY_CACHE=$vcpkgBinaryCache" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append

CMakeLists.txt

@@ -1424,6 +1424,8 @@ if(BUILD_TESTING)
       Int::Arithmetic::Max::Nary
       Int::Cumulative::Man::Fix::0::4
       Int::Distinct::Random
+      Int::Extensional::Reg::Sparse::NFAPrefix
+      Int::Extensional::Reg::Sparse::RandomNFADifferential
       Int::Extensional::TupleSet::Sparse::IncrementalDelta
       Int::Extensional::TupleSet::Auto::DefaultDispatch
       Int::Linear::Bool::Int::Lq

Makefile.in

@@ -1351,6 +1351,8 @@ CHECKTESTS = Branch::Int::Dense::3 \
 	Int::Arithmetic::Max::Nary \
 	Int::Cumulative::Man::Fix::0::4 \
 	Int::Distinct::Random \
+	Int::Extensional::Reg::Sparse::NFAPrefix \
+	Int::Extensional::Reg::Sparse::RandomNFADifferential \
 	Int::Extensional::TupleSet::Sparse::IncrementalDelta \
 	Int::Extensional::TupleSet::Auto::DefaultDispatch \
 	Int::Linear::Bool::Int::Lq \

changelog.in

@@ -67,6 +67,32 @@
 #    optional section in the html page.
 #
 
+[RELEASE]
+Version: 6.5.0
+Date: unreleased
+[DESCRIPTION]
+This is the development changelog for the next Gecode release.
+
+[ENTRY]
+Module: int
+What:   performance
+Rank:   major
+[DESCRIPTION]
+Reduce construction memory for regular constraints by building the layered
+graph from sparse forward frontiers and retaining only backward co-reachable
+states. This avoids allocating one temporary state for every automaton state
+at every sequence position while preserving the persistent graph.
+
+[ENTRY]
+Module: int
+What:   new
+Rank:   major
+[DESCRIPTION]
+Add DFA::nfa factories for finite automata with nondeterministic transitions.
+The factories retain nondeterminism without determinization or minimization;
+epsilon transitions are not supported. Regular propagation now preserves
+branching assigned prefixes when cloning such automata.
+
 [RELEASE]
 Version: 6.4.0
 Date: 2026-07-15

gecode/int.hh

@@ -2192,7 +2192,12 @@ namespace Gecode {
    */
 
   /**
-   * \brief Deterministic finite automaton (%DFA)
+   * \brief Finite automaton with deterministic construction by default
+   *
+   * The constructors and init() require deterministic transitions.
+   * Use nfa() to retain nondeterministic transitions without determinization
+   * or minimization. Every transition consumes one symbol; epsilon
+   * transitions are not supported.
    *
    * After initialization, the start state is always zero.
    * The final states are contiguous ranging from the first to the
@@ -2301,6 +2306,28 @@ namespace Gecode {
     GECODE_INT_EXPORT
     DFA(int s, std::initializer_list<Transition> t,
         std::initializer_list<int> f, bool minimize=true);
+    /**
+     * \brief Construct a nondeterministic finite automaton
+     *
+     * Start state is \a s. The last transition in \a t must have -1 as
+     * its input state, and the last final state in \a f must be -1.
+     * Multiple transitions may share an input state and symbol.
+     * Every transition consumes one symbol; epsilon transitions are not
+     * supported. Unreachable states are removed, but the automaton is
+     * neither determinized nor minimized.
+     */
+    GECODE_INT_EXPORT
+    static DFA nfa(int s, Transition t[], int f[]);
+    /**
+     * \brief Construct a nondeterministic finite automaton from lists
+     *
+     * Start state is \a s, transitions are \a t, and final states are
+     * \a f. No sentinel elements are needed. As for the array overload,
+     * nondeterminism is retained and epsilon transitions are not supported.
+     */
+    GECODE_INT_EXPORT
+    static DFA nfa(int s, std::initializer_list<Transition> t,
+                   std::initializer_list<int> f);
     /// Initialize by DFA \a d (DFA is shared)
     DFA(const DFA& d);
     /// Test whether DFA is equal to \a d
@@ -2717,7 +2744,7 @@ namespace Gecode {
    * \brief Post domain consistent propagator for extensional constraint described by a DFA
    *
    * The elements of \a x must be a word of the language described by
-   * the DFA \a d.
+   * the automaton \a d, which may be constructed with DFA::nfa().
    *
    * Throws an exception of type Int::ArgumentSame, if \a x contains
    * the same unassigned variable multiply. If shared occurrences of variables
@@ -2733,7 +2760,7 @@ namespace Gecode {
    * \brief Post domain consistent propagator for extensional constraint described by a DFA
    *
    * The elements of \a x must be a word of the language described by
-   * the DFA \a d.
+   * the automaton \a d, which may be constructed with DFA::nfa().
    *
    * Throws an exception of type Int::ArgumentSame, if \a x contains
    * the same unassigned variable multiply. If shared occurrences of variables

gecode/int/extensional/dfa.cpp

@@ -497,6 +497,17 @@ namespace Gecode {
     init(start,ts,fs,minimize);
   }
 
+  DFA
+  DFA::nfa(int start, Transition t[], int f[]) {
+    return DFA(start,t,f,false);
+  }
+
+  DFA
+  DFA::nfa(int start, std::initializer_list<Transition> t,
+           std::initializer_list<int> f) {
+    return DFA(start,t,f,false);
+  }
+
   bool
   DFA::equal(const DFA& d) const {
     assert(n_states() == d.n_states());