]> www.ginac.de Git - ginac.git/blobdiff - ginac/normal.cpp
Speed up special cases of square-free factorization.
[ginac.git] / ginac / normal.cpp
index 9ec7574ad665eb33775923238b132c84b29daf9b..9f8b7b4a406fdb5fe5ee39454e86eb6340f0d86a 100644 (file)
@@ -6,7 +6,7 @@
  *  computation, square-free factorization and rational function normalization. */
 
 /*
- *  GiNaC Copyright (C) 1999-2008 Johannes Gutenberg University Mainz, Germany
+ *  GiNaC Copyright (C) 1999-2018 Johannes Gutenberg University Mainz, Germany
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
@@ -23,9 +23,6 @@
  *  Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
  */
 
-#include <algorithm>
-#include <map>
-
 #include "normal.h"
 #include "basic.h"
 #include "ex.h"
 #include "pseries.h"
 #include "symbol.h"
 #include "utils.h"
+#include "polynomial/chinrem_gcd.h"
+
+#include <algorithm>
+#include <map>
 
 namespace GiNaC {
 
@@ -119,6 +120,11 @@ static bool get_first_symbol(const ex &e, ex &x)
  *
  *  @see get_symbol_stats */
 struct sym_desc {
+       /** Initialize symbol, leave other variables uninitialized */
+       sym_desc(const ex& s)
+         : sym(s), deg_a(0), deg_b(0), ldeg_a(0), ldeg_b(0), max_deg(0), max_lcnops(0)
+       { }
+
        /** Reference to symbol */
        ex sym;
 
@@ -140,7 +146,7 @@ struct sym_desc {
        /** Maximum number of terms of leading coefficient of symbol in both polynomials */
        size_t max_lcnops;
 
-       /** Commparison operator for sorting */
+       /** Comparison operator for sorting */
        bool operator<(const sym_desc &x) const
        {
                if (max_deg == x.max_deg)
@@ -156,15 +162,11 @@ typedef std::vector<sym_desc> sym_desc_vec;
 // Add symbol the sym_desc_vec (used internally by get_symbol_stats())
 static void add_symbol(const ex &s, sym_desc_vec &v)
 {
-       sym_desc_vec::const_iterator it = v.begin(), itend = v.end();
-       while (it != itend) {
-               if (it->sym.is_equal(s))  // If it's already in there, don't add it a second time
+       for (auto & it : v)
+               if (it.sym.is_equal(s))  // If it's already in there, don't add it a second time
                        return;
-               ++it;
-       }
-       sym_desc d;
-       d.sym = s;
-       v.push_back(d);
+
+       v.push_back(sym_desc(s));
 }
 
 // Collect all symbols of an expression (used internally by get_symbol_stats())
@@ -194,19 +196,17 @@ static void collect_symbols(const ex &e, sym_desc_vec &v)
  *  @param v  vector of sym_desc structs (filled in) */
 static void get_symbol_stats(const ex &a, const ex &b, sym_desc_vec &v)
 {
-       collect_symbols(a.eval(), v);   // eval() to expand assigned symbols
-       collect_symbols(b.eval(), v);
-       sym_desc_vec::iterator it = v.begin(), itend = v.end();
-       while (it != itend) {
-               int deg_a = a.degree(it->sym);
-               int deg_b = b.degree(it->sym);
-               it->deg_a = deg_a;
-               it->deg_b = deg_b;
-               it->max_deg = std::max(deg_a, deg_b);
-               it->max_lcnops = std::max(a.lcoeff(it->sym).nops(), b.lcoeff(it->sym).nops());
-               it->ldeg_a = a.ldegree(it->sym);
-               it->ldeg_b = b.ldegree(it->sym);
-               ++it;
+       collect_symbols(a, v);
+       collect_symbols(b, v);
+       for (auto & it : v) {
+               int deg_a = a.degree(it.sym);
+               int deg_b = b.degree(it.sym);
+               it.deg_a = deg_a;
+               it.deg_b = deg_b;
+               it.max_deg = std::max(deg_a, deg_b);
+               it.max_lcnops = std::max(a.lcoeff(it.sym).nops(), b.lcoeff(it.sym).nops());
+               it.ldeg_a = a.ldegree(it.sym);
+               it.ldeg_b = b.ldegree(it.sym);
        }
        std::sort(v.begin(), v.end());
 
@@ -270,9 +270,15 @@ static numeric lcm_of_coefficients_denominators(const ex &e)
  *  @param lcm  LCM to multiply in */
 static ex multiply_lcm(const ex &e, const numeric &lcm)
 {
+       if (lcm.is_equal(*_num1_p))
+               // e * 1 -> e;
+               return e;
+
        if (is_exactly_a<mul>(e)) {
+               // (a*b*...)*lcm -> (a*lcma)*(b*lcmb)*...*(lcm/(lcma*lcmb*...))
                size_t num = e.nops();
-               exvector v; v.reserve(num + 1);
+               exvector v;
+               v.reserve(num + 1);
                numeric lcm_accum = *_num1_p;
                for (size_t i=0; i<num; i++) {
                        numeric op_lcm = lcmcoeff(e.op(i), *_num1_p);
@@ -280,20 +286,26 @@ static ex multiply_lcm(const ex &e, const numeric &lcm)
                        lcm_accum *= op_lcm;
                }
                v.push_back(lcm / lcm_accum);
-               return (new mul(v))->setflag(status_flags::dynallocated);
+               return dynallocate<mul>(v);
        } else if (is_exactly_a<add>(e)) {
+               // (a+b+...)*lcm -> a*lcm+b*lcm+...
                size_t num = e.nops();
-               exvector v; v.reserve(num);
+               exvector v;
+               v.reserve(num);
                for (size_t i=0; i<num; i++)
                        v.push_back(multiply_lcm(e.op(i), lcm));
-               return (new add(v))->setflag(status_flags::dynallocated);
+               return dynallocate<add>(v);
        } else if (is_exactly_a<power>(e)) {
-               if (is_a<symbol>(e.op(0)))
-                       return e * lcm;
-               else
-                       return pow(multiply_lcm(e.op(0), lcm.power(ex_to<numeric>(e.op(1)).inverse())), e.op(1));
-       } else
-               return e * lcm;
+               if (!is_a<symbol>(e.op(0))) {
+                       // (b^e)*lcm -> (b*lcm^(1/e))^e if lcm^(1/e) ∈ ℚ (i.e. not a float)
+                       // but not for symbolic b, as evaluation would undo this again
+                       numeric root_of_lcm = lcm.power(ex_to<numeric>(e.op(1)).inverse());
+                       if (root_of_lcm.is_rational())
+                               return pow(multiply_lcm(e.op(0), root_of_lcm), e.op(1));
+               }
+       }
+       // can't recurse down into e
+       return dynallocate<mul>(e, lcm);
 }
 
 
@@ -320,15 +332,12 @@ numeric numeric::integer_content() const
 
 numeric add::integer_content() const
 {
-       epvector::const_iterator it = seq.begin();
-       epvector::const_iterator itend = seq.end();
        numeric c = *_num0_p, l = *_num1_p;
-       while (it != itend) {
-               GINAC_ASSERT(!is_exactly_a<numeric>(it->rest));
-               GINAC_ASSERT(is_exactly_a<numeric>(it->coeff));
-               c = gcd(ex_to<numeric>(it->coeff).numer(), c);
-               l = lcm(ex_to<numeric>(it->coeff).denom(), l);
-               it++;
+       for (auto & it : seq) {
+               GINAC_ASSERT(!is_exactly_a<numeric>(it.rest));
+               GINAC_ASSERT(is_exactly_a<numeric>(it.coeff));
+               c = gcd(ex_to<numeric>(it.coeff).numer(), c);
+               l = lcm(ex_to<numeric>(it.coeff).denom(), l);
        }
        GINAC_ASSERT(is_exactly_a<numeric>(overall_coeff));
        c = gcd(ex_to<numeric>(overall_coeff).numer(), c);
@@ -339,11 +348,8 @@ numeric add::integer_content() const
 numeric mul::integer_content() const
 {
 #ifdef DO_GINAC_ASSERT
-       epvector::const_iterator it = seq.begin();
-       epvector::const_iterator itend = seq.end();
-       while (it != itend) {
-               GINAC_ASSERT(!is_exactly_a<numeric>(recombine_pair_to_ex(*it)));
-               ++it;
+       for (auto & it : seq) {
+               GINAC_ASSERT(!is_exactly_a<numeric>(recombine_pair_to_ex(it)));
        }
 #endif // def DO_GINAC_ASSERT
        GINAC_ASSERT(is_exactly_a<numeric>(overall_coeff));
@@ -392,16 +398,16 @@ ex quo(const ex &a, const ex &b, const ex &x, bool check_args)
                        term = rcoeff / blcoeff;
                else {
                        if (!divide(rcoeff, blcoeff, term, false))
-                               return (new fail())->setflag(status_flags::dynallocated);
+                               return dynallocate<fail>();
                }
-               term *= power(x, rdeg - bdeg);
+               term *= pow(x, rdeg - bdeg);
                v.push_back(term);
                r -= (term * b).expand();
                if (r.is_zero())
                        break;
                rdeg = r.degree(x);
        }
-       return (new add(v))->setflag(status_flags::dynallocated);
+       return dynallocate<add>(v);
 }
 
 
@@ -445,9 +451,9 @@ ex rem(const ex &a, const ex &b, const ex &x, bool check_args)
                        term = rcoeff / blcoeff;
                else {
                        if (!divide(rcoeff, blcoeff, term, false))
-                               return (new fail())->setflag(status_flags::dynallocated);
+                               return dynallocate<fail>();
                }
-               term *= power(x, rdeg - bdeg);
+               term *= pow(x, rdeg - bdeg);
                r -= (term * b).expand();
                if (r.is_zero())
                        break;
@@ -507,23 +513,23 @@ ex prem(const ex &a, const ex &b, const ex &x, bool check_args)
                if (bdeg == 0)
                        eb = _ex0;
                else
-                       eb -= blcoeff * power(x, bdeg);
+                       eb -= blcoeff * pow(x, bdeg);
        } else
                blcoeff = _ex1;
 
        int delta = rdeg - bdeg + 1, i = 0;
        while (rdeg >= bdeg && !r.is_zero()) {
                ex rlcoeff = r.coeff(x, rdeg);
-               ex term = (power(x, rdeg - bdeg) * eb * rlcoeff).expand();
+               ex term = (pow(x, rdeg - bdeg) * eb * rlcoeff).expand();
                if (rdeg == 0)
                        r = _ex0;
                else
-                       r -= rlcoeff * power(x, rdeg);
+                       r -= rlcoeff * pow(x, rdeg);
                r = (blcoeff * r).expand() - term;
                rdeg = r.degree(x);
                i++;
        }
-       return power(blcoeff, delta - i) * r;
+       return pow(blcoeff, delta - i) * r;
 }
 
 
@@ -559,17 +565,17 @@ ex sprem(const ex &a, const ex &b, const ex &x, bool check_args)
                if (bdeg == 0)
                        eb = _ex0;
                else
-                       eb -= blcoeff * power(x, bdeg);
+                       eb -= blcoeff * pow(x, bdeg);
        } else
                blcoeff = _ex1;
 
        while (rdeg >= bdeg && !r.is_zero()) {
                ex rlcoeff = r.coeff(x, rdeg);
-               ex term = (power(x, rdeg - bdeg) * eb * rlcoeff).expand();
+               ex term = (pow(x, rdeg - bdeg) * eb * rlcoeff).expand();
                if (rdeg == 0)
                        r = _ex0;
                else
-                       r -= rlcoeff * power(x, rdeg);
+                       r -= rlcoeff * pow(x, rdeg);
                r = (blcoeff * r).expand() - term;
                rdeg = r.degree(x);
        }
@@ -659,7 +665,7 @@ bool divide(const ex &a, const ex &b, ex &q, bool check_args)
                                else
                                        resv.push_back(a.op(j));
                        }
-                       q = (new mul(resv))->setflag(status_flags::dynallocated);
+                       q = dynallocate<mul>(resv);
                        return true;
                }
        } else if (is_exactly_a<power>(a)) {
@@ -669,15 +675,16 @@ bool divide(const ex &a, const ex &b, ex &q, bool check_args)
                int a_exp = ex_to<numeric>(a.op(1)).to_int();
                ex rem_i;
                if (divide(ab, b, rem_i, false)) {
-                       q = rem_i*power(ab, a_exp - 1);
+                       q = rem_i * pow(ab, a_exp - 1);
                        return true;
                }
-               for (int i=2; i < a_exp; i++) {
-                       if (divide(power(ab, i), b, rem_i, false)) {
-                               q = rem_i*power(ab, a_exp - i);
-                               return true;
-                       }
-               } // ... so we *really* need to expand expression.
+// code below is commented-out because it leads to a significant slowdown
+//             for (int i=2; i < a_exp; i++) {
+//                     if (divide(power(ab, i), b, rem_i, false)) {
+//                             q = rem_i*power(ab, a_exp - i);
+//                             return true;
+//                     }
+//             } // ... so we *really* need to expand expression.
        }
        
        // Polynomial long division (recursive)
@@ -698,11 +705,11 @@ bool divide(const ex &a, const ex &b, ex &q, bool check_args)
                else
                        if (!divide(rcoeff, blcoeff, term, false))
                                return false;
-               term *= power(x, rdeg - bdeg);
+               term *= pow(x, rdeg - bdeg);
                v.push_back(term);
                r -= (term * b).expand();
                if (r.is_zero()) {
-                       q = (new add(v))->setflag(status_flags::dynallocated);
+                       q = dynallocate<add>(v);
                        return true;
                }
                rdeg = r.degree(x);
@@ -794,10 +801,10 @@ static bool divide_in_z(const ex &a, const ex &b, ex &q, sym_desc_vec::const_ite
 
        if (is_exactly_a<mul>(b)) {
                ex qbar = a;
-               for (const_iterator itrb = b.begin(); itrb != b.end(); ++itrb) {
+               for (const auto & it : b) {
                        sym_desc_vec sym_stats;
-                       get_symbol_stats(a, *itrb, sym_stats);
-                       if (!divide_in_z(qbar, *itrb, q, sym_stats.begin()))
+                       get_symbol_stats(a, it, sym_stats);
+                       if (!divide_in_z(qbar, it, q, sym_stats.begin()))
                                return false;
 
                        qbar = q;
@@ -881,11 +888,11 @@ static bool divide_in_z(const ex &a, const ex &b, ex &q, sym_desc_vec::const_ite
                ex term, rcoeff = r.coeff(x, rdeg);
                if (!divide_in_z(rcoeff, blcoeff, term, var+1))
                        break;
-               term = (term * power(x, rdeg - bdeg)).expand();
+               term = (term * pow(x, rdeg - bdeg)).expand();
                v.push_back(term);
                r -= (term * eb).expand();
                if (r.is_zero()) {
-                       q = (new add(v))->setflag(status_flags::dynallocated);
+                       q = dynallocate<add>(v);
 #if USE_REMEMBER
                        dr_remember[ex2(a, b)] = exbool(q, true);
 #endif
@@ -959,7 +966,7 @@ ex ex::content(const ex &x) const
                return lcoeff * c / lcoeff.unit(x);
        ex cont = _ex0;
        for (int i=ldeg; i<=deg; i++)
-               cont = gcd(r.coeff(x, i), cont, NULL, NULL, false);
+               cont = gcd(r.coeff(x, i), cont, nullptr, nullptr, false);
        return cont * c;
 }
 
@@ -1099,7 +1106,7 @@ static ex sr_gcd(const ex &a, const ex &b, sym_desc_vec::const_iterator var)
        // Remove content from c and d, to be attached to GCD later
        ex cont_c = c.content(x);
        ex cont_d = d.content(x);
-       ex gamma = gcd(cont_c, cont_d, NULL, NULL, false);
+       ex gamma = gcd(cont_c, cont_d, nullptr, nullptr, false);
        if (ddeg == 0)
                return gamma;
        c = c.primpart(x, cont_c);
@@ -1163,17 +1170,14 @@ numeric numeric::max_coefficient() const
 
 numeric add::max_coefficient() const
 {
-       epvector::const_iterator it = seq.begin();
-       epvector::const_iterator itend = seq.end();
        GINAC_ASSERT(is_exactly_a<numeric>(overall_coeff));
        numeric cur_max = abs(ex_to<numeric>(overall_coeff));
-       while (it != itend) {
+       for (auto & it : seq) {
                numeric a;
-               GINAC_ASSERT(!is_exactly_a<numeric>(it->rest));
-               a = abs(ex_to<numeric>(it->coeff));
+               GINAC_ASSERT(!is_exactly_a<numeric>(it.rest));
+               a = abs(ex_to<numeric>(it.coeff));
                if (a > cur_max)
                        cur_max = a;
-               it++;
        }
        return cur_max;
 }
@@ -1181,11 +1185,8 @@ numeric add::max_coefficient() const
 numeric mul::max_coefficient() const
 {
 #ifdef DO_GINAC_ASSERT
-       epvector::const_iterator it = seq.begin();
-       epvector::const_iterator itend = seq.end();
-       while (it != itend) {
-               GINAC_ASSERT(!is_exactly_a<numeric>(recombine_pair_to_ex(*it)));
-               it++;
+       for (auto & it : seq) {
+               GINAC_ASSERT(!is_exactly_a<numeric>(recombine_pair_to_ex(it)));
        }
 #endif // def DO_GINAC_ASSERT
        GINAC_ASSERT(is_exactly_a<numeric>(overall_coeff));
@@ -1213,36 +1214,30 @@ ex add::smod(const numeric &xi) const
 {
        epvector newseq;
        newseq.reserve(seq.size()+1);
-       epvector::const_iterator it = seq.begin();
-       epvector::const_iterator itend = seq.end();
-       while (it != itend) {
-               GINAC_ASSERT(!is_exactly_a<numeric>(it->rest));
-               numeric coeff = GiNaC::smod(ex_to<numeric>(it->coeff), xi);
+       for (auto & it : seq) {
+               GINAC_ASSERT(!is_exactly_a<numeric>(it.rest));
+               numeric coeff = GiNaC::smod(ex_to<numeric>(it.coeff), xi);
                if (!coeff.is_zero())
-                       newseq.push_back(expair(it->rest, coeff));
-               it++;
+                       newseq.push_back(expair(it.rest, coeff));
        }
        GINAC_ASSERT(is_exactly_a<numeric>(overall_coeff));
        numeric coeff = GiNaC::smod(ex_to<numeric>(overall_coeff), xi);
-       return (new add(newseq,coeff))->setflag(status_flags::dynallocated);
+       return dynallocate<add>(std::move(newseq), coeff);
 }
 
 ex mul::smod(const numeric &xi) const
 {
 #ifdef DO_GINAC_ASSERT
-       epvector::const_iterator it = seq.begin();
-       epvector::const_iterator itend = seq.end();
-       while (it != itend) {
-               GINAC_ASSERT(!is_exactly_a<numeric>(recombine_pair_to_ex(*it)));
-               it++;
+       for (auto & it : seq) {
+               GINAC_ASSERT(!is_exactly_a<numeric>(recombine_pair_to_ex(it)));
        }
 #endif // def DO_GINAC_ASSERT
-       mul * mulcopyp = new mul(*this);
+       mul & mulcopy = dynallocate<mul>(*this);
        GINAC_ASSERT(is_exactly_a<numeric>(overall_coeff));
-       mulcopyp->overall_coeff = GiNaC::smod(ex_to<numeric>(overall_coeff),xi);
-       mulcopyp->clearflag(status_flags::evaluated);
-       mulcopyp->clearflag(status_flags::hash_calculated);
-       return mulcopyp->setflag(status_flags::dynallocated);
+       mulcopy.overall_coeff = GiNaC::smod(ex_to<numeric>(overall_coeff),xi);
+       mulcopy.clearflag(status_flags::evaluated);
+       mulcopy.clearflag(status_flags::hash_calculated);
+       return mulcopy;
 }
 
 
@@ -1254,10 +1249,10 @@ static ex interpolate(const ex &gamma, const numeric &xi, const ex &x, int degre
        numeric rxi = xi.inverse();
        for (int i=0; !e.is_zero(); i++) {
                ex gi = e.smod(xi);
-               g.push_back(gi * power(x, i));
+               g.push_back(gi * pow(x, i));
                e = (e - gi) * rxi;
        }
-       return (new add(g))->setflag(status_flags::dynallocated);
+       return dynallocate<add>(g);
 }
 
 /** Exception thrown by heur_gcd() to signal failure. */
@@ -1270,9 +1265,9 @@ class gcdheu_failed {};
  *
  *  @param a  first integer multivariate polynomial (expanded)
  *  @param b  second integer multivariate polynomial (expanded)
- *  @param ca  cofactor of polynomial a (returned), NULL to suppress
+ *  @param ca  cofactor of polynomial a (returned), nullptr to suppress
  *             calculation of cofactor
- *  @param cb  cofactor of polynomial b (returned), NULL to suppress
+ *  @param cb  cofactor of polynomial b (returned), nullptr to suppress
  *             calculation of cofactor
  *  @param var iterator to first element of vector of sym_desc structs
  *  @param res the GCD (returned)
@@ -1363,9 +1358,9 @@ static bool heur_gcd_z(ex& res, const ex &a, const ex &b, ex *ca, ex *cb,
  *
  *  @param a  first rational multivariate polynomial (expanded)
  *  @param b  second rational multivariate polynomial (expanded)
- *  @param ca  cofactor of polynomial a (returned), NULL to suppress
+ *  @param ca  cofactor of polynomial a (returned), nullptr to suppress
  *             calculation of cofactor
- *  @param cb  cofactor of polynomial b (returned), NULL to suppress
+ *  @param cb  cofactor of polynomial b (returned), nullptr to suppress
  *             calculation of cofactor
  *  @param var iterator to first element of vector of sym_desc structs
  *  @param res the GCD (returned)
@@ -1415,14 +1410,22 @@ static bool heur_gcd(ex& res, const ex& a, const ex& b, ex *ca, ex *cb,
 }
 
 
+// gcd helper to handle partially factored polynomials (to avoid expanding
+// large expressions). At least one of the arguments should be a power.
+static ex gcd_pf_pow(const ex& a, const ex& b, ex* ca, ex* cb);
+
+// gcd helper to handle partially factored polynomials (to avoid expanding
+// large expressions). At least one of the arguments should be a product.
+static ex gcd_pf_mul(const ex& a, const ex& b, ex* ca, ex* cb);
+
 /** Compute GCD (Greatest Common Divisor) of multivariate polynomials a(X)
  *  and b(X) in Z[X]. Optionally also compute the cofactors of a and b,
  *  defined by a = ca * gcd(a, b) and b = cb * gcd(a, b).
  *
  *  @param a  first multivariate polynomial
  *  @param b  second multivariate polynomial
- *  @param ca pointer to expression that will receive the cofactor of a, or NULL
- *  @param cb pointer to expression that will receive the cofactor of b, or NULL
+ *  @param ca pointer to expression that will receive the cofactor of a, or nullptr
+ *  @param cb pointer to expression that will receive the cofactor of b, or nullptr
  *  @param check_args  check whether a and b are polynomials with rational
  *         coefficients (defaults to "true")
  *  @return the GCD as a new expression */
@@ -1457,150 +1460,14 @@ ex gcd(const ex &a, const ex &b, ex *ca, ex *cb, bool check_args, unsigned optio
        }
 
        // Partially factored cases (to avoid expanding large expressions)
-       if (is_exactly_a<mul>(a)) {
-               if (is_exactly_a<mul>(b) && b.nops() > a.nops())
-                       goto factored_b;
-factored_a:
-               size_t num = a.nops();
-               exvector g; g.reserve(num);
-               exvector acc_ca; acc_ca.reserve(num);
-               ex part_b = b;
-               for (size_t i=0; i<num; i++) {
-                       ex part_ca, part_cb;
-                       g.push_back(gcd(a.op(i), part_b, &part_ca, &part_cb, check_args));
-                       acc_ca.push_back(part_ca);
-                       part_b = part_cb;
-               }
-               if (ca)
-                       *ca = (new mul(acc_ca))->setflag(status_flags::dynallocated);
-               if (cb)
-                       *cb = part_b;
-               return (new mul(g))->setflag(status_flags::dynallocated);
-       } else if (is_exactly_a<mul>(b)) {
-               if (is_exactly_a<mul>(a) && a.nops() > b.nops())
-                       goto factored_a;
-factored_b:
-               size_t num = b.nops();
-               exvector g; g.reserve(num);
-               exvector acc_cb; acc_cb.reserve(num);
-               ex part_a = a;
-               for (size_t i=0; i<num; i++) {
-                       ex part_ca, part_cb;
-                       g.push_back(gcd(part_a, b.op(i), &part_ca, &part_cb, check_args));
-                       acc_cb.push_back(part_cb);
-                       part_a = part_ca;
-               }
-               if (ca)
-                       *ca = part_a;
-               if (cb)
-                       *cb = (new mul(acc_cb))->setflag(status_flags::dynallocated);
-               return (new mul(g))->setflag(status_flags::dynallocated);
-       }
-
+       if (!(options & gcd_options::no_part_factored)) {
+               if (is_exactly_a<mul>(a) || is_exactly_a<mul>(b))
+                       return gcd_pf_mul(a, b, ca, cb);
 #if FAST_COMPARE
-       // Input polynomials of the form poly^n are sometimes also trivial
-       if (is_exactly_a<power>(a)) {
-               ex p = a.op(0);
-               const ex& exp_a = a.op(1);
-               if (is_exactly_a<power>(b)) {
-                       ex pb = b.op(0);
-                       const ex& exp_b = b.op(1);
-                       if (p.is_equal(pb)) {
-                               // a = p^n, b = p^m, gcd = p^min(n, m)
-                               if (exp_a < exp_b) {
-                                       if (ca)
-                                               *ca = _ex1;
-                                       if (cb)
-                                               *cb = power(p, exp_b - exp_a);
-                                       return power(p, exp_a);
-                               } else {
-                                       if (ca)
-                                               *ca = power(p, exp_a - exp_b);
-                                       if (cb)
-                                               *cb = _ex1;
-                                       return power(p, exp_b);
-                               }
-                       } else {
-                               ex p_co, pb_co;
-                               ex p_gcd = gcd(p, pb, &p_co, &pb_co, check_args);
-                               if (p_gcd.is_equal(_ex1)) {
-                                       // a(x) = p(x)^n, b(x) = p_b(x)^m, gcd (p, p_b) = 1 ==>
-                                       // gcd(a,b) = 1
-                                       if (ca)
-                                               *ca = a;
-                                       if (cb)
-                                               *cb = b;
-                                       return _ex1;
-                                       // XXX: do I need to check for p_gcd = -1?
-                               } else {
-                                       // there are common factors:
-                                       // a(x) = g(x)^n A(x)^n, b(x) = g(x)^m B(x)^m ==>
-                                       // gcd(a, b) = g(x)^n gcd(A(x)^n, g(x)^(n-m) B(x)^m
-                                       if (exp_a < exp_b) {
-                                               return power(p_gcd, exp_a)*
-                                                       gcd(power(p_co, exp_a), power(p_gcd, exp_b-exp_a)*power(pb_co, exp_b), ca, cb, false);
-                                       } else {
-                                               return power(p_gcd, exp_b)*
-                                                       gcd(power(p_gcd, exp_a - exp_b)*power(p_co, exp_a), power(pb_co, exp_b), ca, cb, false);
-                                       }
-                               } // p_gcd.is_equal(_ex1)
-                       } // p.is_equal(pb)
-
-               } else {
-                       if (p.is_equal(b)) {
-                               // a = p^n, b = p, gcd = p
-                               if (ca)
-                                       *ca = power(p, a.op(1) - 1);
-                               if (cb)
-                                       *cb = _ex1;
-                               return p;
-                       } 
-
-                       ex p_co, bpart_co;
-                       ex p_gcd = gcd(p, b, &p_co, &bpart_co, false);
-
-                       if (p_gcd.is_equal(_ex1)) {
-                               // a(x) = p(x)^n, gcd(p, b) = 1 ==> gcd(a, b) = 1
-                               if (ca)
-                                       *ca = a;
-                               if (cb)
-                                       *cb = b;
-                               return _ex1;
-                       } else {
-                               // a(x) = g(x)^n A(x)^n, b(x) = g(x) B(x) ==> gcd(a, b) = g(x) gcd(g(x)^(n-1) A(x)^n, B(x))
-                               return p_gcd*gcd(power(p_gcd, exp_a-1)*power(p_co, exp_a), bpart_co, ca, cb, false);
-                       }
-               } // is_exactly_a<power>(b)
-
-       } else if (is_exactly_a<power>(b)) {
-               ex p = b.op(0);
-               if (p.is_equal(a)) {
-                       // a = p, b = p^n, gcd = p
-                       if (ca)
-                               *ca = _ex1;
-                       if (cb)
-                               *cb = power(p, b.op(1) - 1);
-                       return p;
-               }
-
-               ex p_co, apart_co;
-               const ex& exp_b(b.op(1));
-               ex p_gcd = gcd(a, p, &apart_co, &p_co, false);
-               if (p_gcd.is_equal(_ex1)) {
-                       // b=p(x)^n, gcd(a, p) = 1 ==> gcd(a, b) == 1
-                       if (ca)
-                               *ca = a;
-                       if (cb)
-                               *cb = b;
-                       return _ex1;
-               } else {
-                       // there are common factors:
-                       // a(x) = g(x) A(x), b(x) = g(x)^n B(x)^n ==> gcd = g(x) gcd(g(x)^(n-1) A(x)^n, B(x))
-
-                       return p_gcd*gcd(apart_co, power(p_gcd, exp_b-1)*power(p_co, exp_b), ca, cb, false);
-               } // p_gcd.is_equal(_ex1)
-       }
+               if (is_exactly_a<power>(a) || is_exactly_a<power>(b))
+                       return gcd_pf_pow(a, b, ca, cb);
 #endif
+       }
 
        // Some trivial cases
        ex aex = a.expand(), bex = b.expand();
@@ -1661,7 +1528,7 @@ factored_b:
                if (ca)
                        *ca = ex_to<numeric>(aex)/g;
                if (cb)
-                       *cb = bex/g;
+                       *cb = bex/g;
                return g;
        }
 
@@ -1681,7 +1548,7 @@ factored_b:
 
        // The symbol with least degree which is contained in both polynomials
        // is our main variable
-       sym_desc_vec::iterator vari = sym_stats.begin();
+       auto vari = sym_stats.begin();
        while ((vari != sym_stats.end()) && 
               (((vari->ldeg_b == 0) && (vari->deg_b == 0)) ||
                ((vari->ldeg_a == 0) && (vari->deg_a == 0))))
@@ -1708,7 +1575,7 @@ factored_b:
        int ldeg_b = var->ldeg_b;
        int min_ldeg = std::min(ldeg_a,ldeg_b);
        if (min_ldeg > 0) {
-               ex common = power(x, min_ldeg);
+               ex common = pow(x, min_ldeg);
                return gcd((aex / common).expand(), (bex / common).expand(), ca, cb, false) * common;
        }
 
@@ -1731,37 +1598,162 @@ factored_b:
 
        // Try heuristic algorithm first, fall back to PRS if that failed
        ex g;
-       bool found = heur_gcd(g, aex, bex, ca, cb, var);
-       if (!found) {
+       if (!(options & gcd_options::no_heur_gcd)) {
+               bool found = heur_gcd(g, aex, bex, ca, cb, var);
+               if (found) {
+                       // heur_gcd have already computed cofactors...
+                       if (g.is_equal(_ex1)) {
+                               // ... but we want to keep them factored if possible.
+                               if (ca)
+                                       *ca = a;
+                               if (cb)
+                                       *cb = b;
+                       }
+                       return g;
+               }
 #if STATISTICS
-               heur_gcd_failed++;
+               else {
+                       heur_gcd_failed++;
+               }
 #endif
+       }
+       if (options & gcd_options::use_sr_gcd) {
                g = sr_gcd(aex, bex, var);
-               if (g.is_equal(_ex1)) {
-                       // Keep cofactors factored if possible
+       } else {
+               exvector vars;
+               for (std::size_t n = sym_stats.size(); n-- != 0; )
+                       vars.push_back(sym_stats[n].sym);
+               g = chinrem_gcd(aex, bex, vars);
+       }
+
+       if (g.is_equal(_ex1)) {
+               // Keep cofactors factored if possible
+               if (ca)
+                       *ca = a;
+               if (cb)
+                       *cb = b;
+       } else {
+               if (ca)
+                       divide(aex, g, *ca, false);
+               if (cb)
+                       divide(bex, g, *cb, false);
+       }
+       return g;
+}
+
+// gcd helper to handle partially factored polynomials (to avoid expanding
+// large expressions). Both arguments should be powers.
+static ex gcd_pf_pow_pow(const ex& a, const ex& b, ex* ca, ex* cb)
+{
+       ex p = a.op(0);
+       const ex& exp_a = a.op(1);
+       ex pb = b.op(0);
+       const ex& exp_b = b.op(1);
+
+       // a = p^n, b = p^m, gcd = p^min(n, m)
+       if (p.is_equal(pb)) {
+               if (exp_a < exp_b) {
                        if (ca)
-                               *ca = a;
+                               *ca = _ex1;
                        if (cb)
-                               *cb = b;
+                               *cb = pow(p, exp_b - exp_a);
+                       return pow(p, exp_a);
                } else {
                        if (ca)
-                               divide(aex, g, *ca, false);
+                               *ca = pow(p, exp_a - exp_b);
                        if (cb)
-                               divide(bex, g, *cb, false);
+                               *cb = _ex1;
+                       return pow(p, exp_b);
                }
-       } else {
-               if (g.is_equal(_ex1)) {
-                       // Keep cofactors factored if possible
+       }
+
+       ex p_co, pb_co;
+       ex p_gcd = gcd(p, pb, &p_co, &pb_co, false);
+       // a(x) = p(x)^n, b(x) = p_b(x)^m, gcd (p, p_b) = 1 ==> gcd(a,b) = 1
+       if (p_gcd.is_equal(_ex1)) {
                        if (ca)
                                *ca = a;
                        if (cb)
                                *cb = b;
-               }
+                       return _ex1;
+                       // XXX: do I need to check for p_gcd = -1?
        }
 
-       return g;
+       // there are common factors:
+       // a(x) = g(x)^n A(x)^n, b(x) = g(x)^m B(x)^m ==>
+       // gcd(a, b) = g(x)^n gcd(A(x)^n, g(x)^(n-m) B(x)^m
+       if (exp_a < exp_b) {
+               ex pg =  gcd(pow(p_co, exp_a), pow(p_gcd, exp_b-exp_a)*pow(pb_co, exp_b), ca, cb, false);
+               return pow(p_gcd, exp_a)*pg;
+       } else {
+               ex pg = gcd(pow(p_gcd, exp_a - exp_b)*pow(p_co, exp_a), pow(pb_co, exp_b), ca, cb, false);
+               return pow(p_gcd, exp_b)*pg;
+       }
 }
 
+static ex gcd_pf_pow(const ex& a, const ex& b, ex* ca, ex* cb)
+{
+       if (is_exactly_a<power>(a) && is_exactly_a<power>(b))
+               return gcd_pf_pow_pow(a, b, ca, cb);
+
+       if (is_exactly_a<power>(b) && (! is_exactly_a<power>(a)))
+               return gcd_pf_pow(b, a, cb, ca);
+
+       GINAC_ASSERT(is_exactly_a<power>(a));
+
+       ex p = a.op(0);
+       const ex& exp_a = a.op(1);
+       if (p.is_equal(b)) {
+               // a = p^n, b = p, gcd = p
+               if (ca)
+                       *ca = pow(p, a.op(1) - 1);
+               if (cb)
+                       *cb = _ex1;
+               return p;
+       } 
+
+       ex p_co, bpart_co;
+       ex p_gcd = gcd(p, b, &p_co, &bpart_co, false);
+
+       // a(x) = p(x)^n, gcd(p, b) = 1 ==> gcd(a, b) = 1
+       if (p_gcd.is_equal(_ex1)) {
+               if (ca)
+                       *ca = a;
+               if (cb)
+                       *cb = b;
+               return _ex1;
+       }
+       // a(x) = g(x)^n A(x)^n, b(x) = g(x) B(x) ==> gcd(a, b) = g(x) gcd(g(x)^(n-1) A(x)^n, B(x))
+       ex rg = gcd(pow(p_gcd, exp_a-1)*pow(p_co, exp_a), bpart_co, ca, cb, false);
+       return p_gcd*rg;
+}
+
+static ex gcd_pf_mul(const ex& a, const ex& b, ex* ca, ex* cb)
+{
+       if (is_exactly_a<mul>(a) && is_exactly_a<mul>(b)
+                                && (b.nops() >  a.nops()))
+               return gcd_pf_mul(b, a, cb, ca);
+
+       if (is_exactly_a<mul>(b) && (!is_exactly_a<mul>(a)))
+               return gcd_pf_mul(b, a, cb, ca);
+
+       GINAC_ASSERT(is_exactly_a<mul>(a));
+       size_t num = a.nops();
+       exvector g; g.reserve(num);
+       exvector acc_ca; acc_ca.reserve(num);
+       ex part_b = b;
+       for (size_t i=0; i<num; i++) {
+               ex part_ca, part_cb;
+               g.push_back(gcd(a.op(i), part_b, &part_ca, &part_cb, false));
+               acc_ca.push_back(part_ca);
+               part_b = part_cb;
+       }
+       if (ca)
+               *ca = dynallocate<mul>(acc_ca);
+       if (cb)
+               *cb = part_b;
+       return dynallocate<mul>(g);
+}
 
 /** Compute LCM (Least Common Multiple) of multivariate polynomials in Z[X].
  *
@@ -1791,34 +1783,47 @@ ex lcm(const ex &a, const ex &b, bool check_args)
  *  Yun's algorithm.  Used internally by sqrfree().
  *
  *  @param a  multivariate polynomial over Z[X], treated here as univariate
- *            polynomial in x.
+ *            polynomial in x (needs not be expanded).
  *  @param x  variable to factor in
- *  @return   vector of factors sorted in ascending degree */
-static exvector sqrfree_yun(const ex &a, const symbol &x)
+ *  @return   vector of expairs (factor, exponent), sorted by exponents */
+static epvector sqrfree_yun(const ex &a, const symbol &x)
 {
-       exvector res;
        ex w = a;
        ex z = w.diff(x);
        ex g = gcd(w, z);
+       if (g.is_zero()) {
+               return epvector{};
+       }
        if (g.is_equal(_ex1)) {
-               res.push_back(a);
-               return res;
+               return epvector{expair(a, _ex1)};
        }
-       ex y;
+       epvector results;
+       ex exponent = _ex0;
        do {
                w = quo(w, g, x);
-               y = quo(z, g, x);
-               z = y - w.diff(x);
+               if (w.is_zero()) {
+                       return res;
+               }
+               z = quo(z, g, x) - w.diff(x);
+               exponent = exponent + 1;
+               if (w.is_equal(x)) {
+                       // shortcut for x^n with n ∈ ℕ
+                       exponent += quo(z, w.diff(x), x);
+                       results.push_back(expair(w, exponent));
+                       break;
+               }
                g = gcd(w, z);
-               res.push_back(g);
+               if (!g.is_equal(_ex1)) {
+                       results.push_back(expair(g, exponent));
+               }
        } while (!z.is_zero());
-       return res;
+       return results;
 }
 
 
 /** Compute a square-free factorization of a multivariate polynomial in Q[X].
  *
- *  @param a  multivariate polynomial over Q[X]
+ *  @param a  multivariate polynomial over Q[X] (needs not be expanded)
  *  @param l  lst of variables to factor in, may be left empty for autodetection
  *  @return   a square-free factorization of \p a.
  *
@@ -1853,8 +1858,8 @@ static exvector sqrfree_yun(const ex &a, const symbol &x)
  */
 ex sqrfree(const ex &a, const lst &l)
 {
-       if (is_exactly_a<numeric>(a) ||     // algorithm does not trap a==0
-           is_a<symbol>(a))        // shortcut
+       if (is_exactly_a<numeric>(a) ||
+           is_a<symbol>(a))        // shortcuts
                return a;
 
        // If no lst of variables to factorize in was specified we have to
@@ -1864,11 +1869,8 @@ ex sqrfree(const ex &a, const lst &l)
        if (l.nops()==0) {
                sym_desc_vec sdv;
                get_symbol_stats(a, _ex0, sdv);
-               sym_desc_vec::const_iterator it = sdv.begin(), itend = sdv.end();
-               while (it != itend) {
-                       args.append(it->sym);
-                       ++it;
-               }
+               for (auto & it : sdv)
+                       args.append(it.sym);
        } else {
                args = l;
        }
@@ -1883,37 +1885,32 @@ ex sqrfree(const ex &a, const lst &l)
        const ex tmp = multiply_lcm(a,lcm);
 
        // find the factors
-       exvector factors = sqrfree_yun(tmp, x);
+       epvector factors = sqrfree_yun(tmp, x);
 
-       // construct the next list of symbols with the first element popped
-       lst newargs = args;
-       newargs.remove_first();
+       // remove symbol x and proceed recursively with the remaining symbols
+       args.remove_first();
 
        // recurse down the factors in remaining variables
-       if (newargs.nops()>0) {
-               exvector::iterator i = factors.begin();
-               while (i != factors.end()) {
-                       *i = sqrfree(*i, newargs);
-                       ++i;
-               }
+       if (args.nops()>0) {
+               for (auto & it : factors)
+                       it.rest = sqrfree(it.rest, args);
        }
 
        // Done with recursion, now construct the final result
        ex result = _ex1;
-       exvector::const_iterator it = factors.begin(), itend = factors.end();
-       for (int p = 1; it!=itend; ++it, ++p)
-               result *= power(*it, p);
+       for (auto & it : factors)
+               result *= pow(it.rest, it.coeff);
 
        // Yun's algorithm does not account for constant factors.  (For univariate
        // polynomials it works only in the monic case.)  We can correct this by
        // inserting what has been lost back into the result.  For completeness
        // we'll also have to recurse down that factor in the remaining variables.
-       if (newargs.nops()>0)
-               result *= sqrfree(quo(tmp, result, x), newargs);
+       if (args.nops()>0)
+               result *= sqrfree(quo(tmp, result, x), args);
        else
                result *= quo(tmp, result, x);
 
-       // Put in the reational overall factor again and return
+       // Put in the rational overall factor again and return
        return result * lcm.inverse();
 }
 
@@ -1937,24 +1934,21 @@ ex sqrfree_parfrac(const ex & a, const symbol & x)
 //clog << "red_poly = " << red_poly << ", red_numer = " << red_numer << endl;
 
        // Factorize denominator and compute cofactors
-       exvector yun = sqrfree_yun(denom, x);
-//clog << "yun factors: " << exprseq(yun) << endl;
-       size_t num_yun = yun.size();
-       exvector factor; factor.reserve(num_yun);
-       exvector cofac; cofac.reserve(num_yun);
-       for (size_t i=0; i<num_yun; i++) {
-               if (!yun[i].is_equal(_ex1)) {
-                       for (size_t j=0; j<=i; j++) {
-                               factor.push_back(pow(yun[i], j+1));
-                               ex prod = _ex1;
-                               for (size_t k=0; k<num_yun; k++) {
-                                       if (k == i)
-                                               prod *= pow(yun[k], i-j);
-                                       else
-                                               prod *= pow(yun[k], k+1);
-                               }
-                               cofac.push_back(prod.expand());
+       epvector yun = sqrfree_yun(denom, x);
+       size_t yun_max_exponent = yun.empty() ? 0 : ex_to<numeric>(yun.back().coeff).to_int();
+       exvector factor, cofac;
+       for (size_t i=0; i<yun.size(); i++) {
+               numeric i_exponent = ex_to<numeric>(yun[i].coeff);
+               for (size_t j=0; j<i_exponent; j++) {
+                       factor.push_back(pow(yun[i].rest, j+1));
+                       ex prod = _ex1;
+                       for (size_t k=0; k<yun.size(); k++) {
+                               if (yun[k].coeff == i_exponent)
+                                       prod *= pow(yun[k].rest, i_exponent-1-j);
+                               else
+                                       prod *= pow(yun[k].rest, yun[k].coeff);
                        }
+                       cofac.push_back(prod.expand());
                }
        }
        size_t num_factors = factor.size();
@@ -2007,16 +2001,18 @@ ex sqrfree_parfrac(const ex & a, const symbol & x)
  *  @see ex::normal */
 static ex replace_with_symbol(const ex & e, exmap & repl, exmap & rev_lookup)
 {
+       // Since the repl contains replaced expressions we should search for them
+       ex e_replaced = e.subs(repl, subs_options::no_pattern);
+
        // Expression already replaced? Then return the assigned symbol
-       exmap::const_iterator it = rev_lookup.find(e);
+       auto it = rev_lookup.find(e_replaced);
        if (it != rev_lookup.end())
                return it->second;
-       
+
        // Otherwise create new symbol and add to list, taking care that the
        // replacement expression doesn't itself contain symbols from repl,
        // because subs() is not recursive
-       ex es = (new symbol)->setflag(status_flags::dynallocated);
-       ex e_replaced = e.subs(repl, subs_options::no_pattern);
+       ex es = dynallocate<symbol>();
        repl.insert(std::make_pair(es, e_replaced));
        rev_lookup.insert(std::make_pair(e_replaced, es));
        return es;
@@ -2029,16 +2025,18 @@ static ex replace_with_symbol(const ex & e, exmap & repl, exmap & rev_lookup)
  *  @see basic::to_polynomial */
 static ex replace_with_symbol(const ex & e, exmap & repl)
 {
+       // Since the repl contains replaced expressions we should search for them
+       ex e_replaced = e.subs(repl, subs_options::no_pattern);
+
        // Expression already replaced? Then return the assigned symbol
-       for (exmap::const_iterator it = repl.begin(); it != repl.end(); ++it)
-               if (it->second.is_equal(e))
-                       return it->first;
-       
+       for (auto & it : repl)
+               if (it.second.is_equal(e_replaced))
+                       return it.first;
+
        // Otherwise create new symbol and add to list, taking care that the
        // replacement expression doesn't itself contain symbols from repl,
        // because subs() is not recursive
-       ex es = (new symbol)->setflag(status_flags::dynallocated);
-       ex e_replaced = e.subs(repl, subs_options::no_pattern);
+       ex es = dynallocate<symbol>();
        repl.insert(std::make_pair(es, e_replaced));
        return es;
 }
@@ -2046,36 +2044,27 @@ static ex replace_with_symbol(const ex & e, exmap & repl)
 
 /** Function object to be applied by basic::normal(). */
 struct normal_map_function : public map_function {
-       int level;
-       normal_map_function(int l) : level(l) {}
-       ex operator()(const ex & e) { return normal(e, level); }
+       ex operator()(const ex & e) override { return normal(e); }
 };
 
 /** Default implementation of ex::normal(). It normalizes the children and
  *  replaces the object with a temporary symbol.
  *  @see ex::normal */
-ex basic::normal(exmap & repl, exmap & rev_lookup, int level) const
+ex basic::normal(exmap & repl, exmap & rev_lookup) const
 {
        if (nops() == 0)
-               return (new lst(replace_with_symbol(*this, repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
-       else {
-               if (level == 1)
-                       return (new lst(replace_with_symbol(*this, repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
-               else if (level == -max_recursion_level)
-                       throw(std::runtime_error("max recursion level reached"));
-               else {
-                       normal_map_function map_normal(level - 1);
-                       return (new lst(replace_with_symbol(map(map_normal), repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
-               }
-       }
+               return dynallocate<lst>({replace_with_symbol(*this, repl, rev_lookup), _ex1});
+
+       normal_map_function map_normal;
+       return dynallocate<lst>({replace_with_symbol(map(map_normal), repl, rev_lookup), _ex1});
 }
 
 
 /** Implementation of ex::normal() for symbols. This returns the unmodified symbol.
  *  @see ex::normal */
-ex symbol::normal(exmap & repl, exmap & rev_lookup, int level) const
+ex symbol::normal(exmap & repl, exmap & rev_lookup) const
 {
-       return (new lst(*this, _ex1))->setflag(status_flags::dynallocated);
+       return dynallocate<lst>({*this, _ex1});
 }
 
 
@@ -2083,7 +2072,7 @@ ex symbol::normal(exmap & repl, exmap & rev_lookup, int level) const
  *  into re+I*im and replaces I and non-rational real numbers with a temporary
  *  symbol.
  *  @see ex::normal */
-ex numeric::normal(exmap & repl, exmap & rev_lookup, int level) const
+ex numeric::normal(exmap & repl, exmap & rev_lookup) const
 {
        numeric num = numer();
        ex numex = num;
@@ -2099,7 +2088,7 @@ ex numeric::normal(exmap & repl, exmap & rev_lookup, int level) const
        }
 
        // Denominator is always a real integer (see numeric::denom())
-       return (new lst(numex, denom()))->setflag(status_flags::dynallocated);
+       return dynallocate<lst>({numex, denom()});
 }
 
 
@@ -2117,11 +2106,11 @@ static ex frac_cancel(const ex &n, const ex &d)
 
        // Handle trivial case where denominator is 1
        if (den.is_equal(_ex1))
-               return (new lst(num, den))->setflag(status_flags::dynallocated);
+               return dynallocate<lst>({num, den});
 
        // Handle special cases where numerator or denominator is 0
        if (num.is_zero())
-               return (new lst(num, _ex1))->setflag(status_flags::dynallocated);
+               return dynallocate<lst>({num, _ex1});
        if (den.expand().is_zero())
                throw(std::overflow_error("frac_cancel: division by zero in frac_cancel"));
 
@@ -2160,32 +2149,25 @@ static ex frac_cancel(const ex &n, const ex &d)
 
        // Return result as list
 //std::clog << " returns num = " << num << ", den = " << den << ", pre_factor = " << pre_factor << std::endl;
-       return (new lst(num * pre_factor.numer(), den * pre_factor.denom()))->setflag(status_flags::dynallocated);
+       return dynallocate<lst>({num * pre_factor.numer(), den * pre_factor.denom()});
 }
 
 
 /** Implementation of ex::normal() for a sum. It expands terms and performs
  *  fractional addition.
  *  @see ex::normal */
-ex add::normal(exmap & repl, exmap & rev_lookup, int level) const
+ex add::normal(exmap & repl, exmap & rev_lookup) const
 {
-       if (level == 1)
-               return (new lst(replace_with_symbol(*this, repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
-       else if (level == -max_recursion_level)
-               throw(std::runtime_error("max recursion level reached"));
-
        // Normalize children and split each one into numerator and denominator
        exvector nums, dens;
        nums.reserve(seq.size()+1);
        dens.reserve(seq.size()+1);
-       epvector::const_iterator it = seq.begin(), itend = seq.end();
-       while (it != itend) {
-               ex n = ex_to<basic>(recombine_pair_to_ex(*it)).normal(repl, rev_lookup, level-1);
+       for (auto & it : seq) {
+               ex n = ex_to<basic>(recombine_pair_to_ex(it)).normal(repl, rev_lookup);
                nums.push_back(n.op(0));
                dens.push_back(n.op(1));
-               it++;
        }
-       ex n = ex_to<numeric>(overall_coeff).normal(repl, rev_lookup, level-1);
+       ex n = ex_to<numeric>(overall_coeff).normal(repl, rev_lookup);
        nums.push_back(n.op(0));
        dens.push_back(n.op(1));
        GINAC_ASSERT(nums.size() == dens.size());
@@ -2195,8 +2177,8 @@ ex add::normal(exmap & repl, exmap & rev_lookup, int level) const
 //std::clog << "add::normal uses " << nums.size() << " summands:\n";
 
        // Add fractions sequentially
-       exvector::const_iterator num_it = nums.begin(), num_itend = nums.end();
-       exvector::const_iterator den_it = dens.begin(), den_itend = dens.end();
+       auto num_it = nums.begin(), num_itend = nums.end();
+       auto den_it = dens.begin(), den_itend = dens.end();
 //std::clog << " num = " << *num_it << ", den = " << *den_it << std::endl;
        ex num = *num_it++, den = *den_it++;
        while (num_it != num_itend) {
@@ -2209,7 +2191,7 @@ ex add::normal(exmap & repl, exmap & rev_lookup, int level) const
                        num_it++; den_it++;
                }
 
-               // Additiion of two fractions, taking advantage of the fact that
+               // Addition of two fractions, taking advantage of the fact that
                // the heuristic GCD algorithm computes the cofactors at no extra cost
                ex co_den1, co_den2;
                ex g = gcd(den, next_den, &co_den1, &co_den2, false);
@@ -2226,31 +2208,23 @@ ex add::normal(exmap & repl, exmap & rev_lookup, int level) const
 /** Implementation of ex::normal() for a product. It cancels common factors
  *  from fractions.
  *  @see ex::normal() */
-ex mul::normal(exmap & repl, exmap & rev_lookup, int level) const
+ex mul::normal(exmap & repl, exmap & rev_lookup) const
 {
-       if (level == 1)
-               return (new lst(replace_with_symbol(*this, repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
-       else if (level == -max_recursion_level)
-               throw(std::runtime_error("max recursion level reached"));
-
        // Normalize children, separate into numerator and denominator
        exvector num; num.reserve(seq.size());
        exvector den; den.reserve(seq.size());
        ex n;
-       epvector::const_iterator it = seq.begin(), itend = seq.end();
-       while (it != itend) {
-               n = ex_to<basic>(recombine_pair_to_ex(*it)).normal(repl, rev_lookup, level-1);
+       for (auto & it : seq) {
+               n = ex_to<basic>(recombine_pair_to_ex(it)).normal(repl, rev_lookup);
                num.push_back(n.op(0));
                den.push_back(n.op(1));
-               it++;
        }
-       n = ex_to<numeric>(overall_coeff).normal(repl, rev_lookup, level-1);
+       n = ex_to<numeric>(overall_coeff).normal(repl, rev_lookup);
        num.push_back(n.op(0));
        den.push_back(n.op(1));
 
        // Perform fraction cancellation
-       return frac_cancel((new mul(num))->setflag(status_flags::dynallocated),
-                          (new mul(den))->setflag(status_flags::dynallocated));
+       return frac_cancel(dynallocate<mul>(num), dynallocate<mul>(den));
 }
 
 
@@ -2258,16 +2232,11 @@ ex mul::normal(exmap & repl, exmap & rev_lookup, int level) const
  *  distributes integer exponents to numerator and denominator, and replaces
  *  non-integer powers by temporary symbols.
  *  @see ex::normal */
-ex power::normal(exmap & repl, exmap & rev_lookup, int level) const
+ex power::normal(exmap & repl, exmap & rev_lookup) const
 {
-       if (level == 1)
-               return (new lst(replace_with_symbol(*this, repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
-       else if (level == -max_recursion_level)
-               throw(std::runtime_error("max recursion level reached"));
-
        // Normalize basis and exponent (exponent gets reassembled)
-       ex n_basis = ex_to<basic>(basis).normal(repl, rev_lookup, level-1);
-       ex n_exponent = ex_to<basic>(exponent).normal(repl, rev_lookup, level-1);
+       ex n_basis = ex_to<basic>(basis).normal(repl, rev_lookup);
+       ex n_exponent = ex_to<basic>(exponent).normal(repl, rev_lookup);
        n_exponent = n_exponent.op(0) / n_exponent.op(1);
 
        if (n_exponent.info(info_flags::integer)) {
@@ -2275,12 +2244,12 @@ ex power::normal(exmap & repl, exmap & rev_lookup, int level) const
                if (n_exponent.info(info_flags::positive)) {
 
                        // (a/b)^n -> {a^n, b^n}
-                       return (new lst(power(n_basis.op(0), n_exponent), power(n_basis.op(1), n_exponent)))->setflag(status_flags::dynallocated);
+                       return dynallocate<lst>({pow(n_basis.op(0), n_exponent), pow(n_basis.op(1), n_exponent)});
 
                } else if (n_exponent.info(info_flags::negative)) {
 
                        // (a/b)^-n -> {b^n, a^n}
-                       return (new lst(power(n_basis.op(1), -n_exponent), power(n_basis.op(0), -n_exponent)))->setflag(status_flags::dynallocated);
+                       return dynallocate<lst>({pow(n_basis.op(1), -n_exponent), pow(n_basis.op(0), -n_exponent)});
                }
 
        } else {
@@ -2288,43 +2257,41 @@ ex power::normal(exmap & repl, exmap & rev_lookup, int level) const
                if (n_exponent.info(info_flags::positive)) {
 
                        // (a/b)^x -> {sym((a/b)^x), 1}
-                       return (new lst(replace_with_symbol(power(n_basis.op(0) / n_basis.op(1), n_exponent), repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
+                       return dynallocate<lst>({replace_with_symbol(pow(n_basis.op(0) / n_basis.op(1), n_exponent), repl, rev_lookup), _ex1});
 
                } else if (n_exponent.info(info_flags::negative)) {
 
                        if (n_basis.op(1).is_equal(_ex1)) {
 
                                // a^-x -> {1, sym(a^x)}
-                               return (new lst(_ex1, replace_with_symbol(power(n_basis.op(0), -n_exponent), repl, rev_lookup)))->setflag(status_flags::dynallocated);
+                               return dynallocate<lst>({_ex1, replace_with_symbol(pow(n_basis.op(0), -n_exponent), repl, rev_lookup)});
 
                        } else {
 
                                // (a/b)^-x -> {sym((b/a)^x), 1}
-                               return (new lst(replace_with_symbol(power(n_basis.op(1) / n_basis.op(0), -n_exponent), repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
+                               return dynallocate<lst>({replace_with_symbol(pow(n_basis.op(1) / n_basis.op(0), -n_exponent), repl, rev_lookup), _ex1});
                        }
                }
        }
 
        // (a/b)^x -> {sym((a/b)^x, 1}
-       return (new lst(replace_with_symbol(power(n_basis.op(0) / n_basis.op(1), n_exponent), repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
+       return dynallocate<lst>({replace_with_symbol(pow(n_basis.op(0) / n_basis.op(1), n_exponent), repl, rev_lookup), _ex1});
 }
 
 
 /** Implementation of ex::normal() for pseries. It normalizes each coefficient
  *  and replaces the series by a temporary symbol.
  *  @see ex::normal */
-ex pseries::normal(exmap & repl, exmap & rev_lookup, int level) const
+ex pseries::normal(exmap & repl, exmap & rev_lookup) const
 {
        epvector newseq;
-       epvector::const_iterator i = seq.begin(), end = seq.end();
-       while (i != end) {
-               ex restexp = i->rest.normal();
+       for (auto & it : seq) {
+               ex restexp = it.rest.normal();
                if (!restexp.is_zero())
-                       newseq.push_back(expair(restexp, i->coeff));
-               ++i;
+                       newseq.push_back(expair(restexp, it.coeff));
        }
-       ex n = pseries(relational(var,point), newseq);
-       return (new lst(replace_with_symbol(n, repl, rev_lookup), _ex1))->setflag(status_flags::dynallocated);
+       ex n = pseries(relational(var,point), std::move(newseq));
+       return dynallocate<lst>({replace_with_symbol(n, repl, rev_lookup), _ex1});
 }
 
 
@@ -2338,13 +2305,12 @@ ex pseries::normal(exmap & repl, exmap & rev_lookup, int level) const
  *  expression can be treated as a rational function). normal() is applied
  *  recursively to arguments of functions etc.
  *
- *  @param level maximum depth of recursion
  *  @return normalized expression */
-ex ex::normal(int level) const
+ex ex::normal() const
 {
        exmap repl, rev_lookup;
 
-       ex e = bp->normal(repl, rev_lookup, level);
+       ex e = bp->normal(repl, rev_lookup);
        GINAC_ASSERT(is_a<lst>(e));
 
        // Re-insert replaced symbols
@@ -2365,7 +2331,7 @@ ex ex::numer() const
 {
        exmap repl, rev_lookup;
 
-       ex e = bp->normal(repl, rev_lookup, 0);
+       ex e = bp->normal(repl, rev_lookup);
        GINAC_ASSERT(is_a<lst>(e));
 
        // Re-insert replaced symbols
@@ -2385,7 +2351,7 @@ ex ex::denom() const
 {
        exmap repl, rev_lookup;
 
-       ex e = bp->normal(repl, rev_lookup, 0);
+       ex e = bp->normal(repl, rev_lookup);
        GINAC_ASSERT(is_a<lst>(e));
 
        // Re-insert replaced symbols
@@ -2395,7 +2361,7 @@ ex ex::denom() const
                return e.op(1).subs(repl, subs_options::no_pattern);
 }
 
-/** Get numerator and denominator of an expression. If the expresison is not
+/** Get numerator and denominator of an expression. If the expression is not
  *  of the normal form "numerator/denominator", it is first converted to this
  *  form and then a list [numerator, denominator] is returned.
  *
@@ -2405,7 +2371,7 @@ ex ex::numer_denom() const
 {
        exmap repl, rev_lookup;
 
-       ex e = bp->normal(repl, rev_lookup, 0);
+       ex e = bp->normal(repl, rev_lookup);
        GINAC_ASSERT(is_a<lst>(e));
 
        // Re-insert replaced symbols
@@ -2434,47 +2400,11 @@ ex ex::to_rational(exmap & repl) const
        return bp->to_rational(repl);
 }
 
-// GiNaC 1.1 compatibility function
-ex ex::to_rational(lst & repl_lst) const
-{
-       // Convert lst to exmap
-       exmap m;
-       for (lst::const_iterator it = repl_lst.begin(); it != repl_lst.end(); ++it)
-               m.insert(std::make_pair(it->op(0), it->op(1)));
-
-       ex ret = bp->to_rational(m);
-
-       // Convert exmap back to lst
-       repl_lst.remove_all();
-       for (exmap::const_iterator it = m.begin(); it != m.end(); ++it)
-               repl_lst.append(it->first == it->second);
-
-       return ret;
-}
-
 ex ex::to_polynomial(exmap & repl) const
 {
        return bp->to_polynomial(repl);
 }
 
-// GiNaC 1.1 compatibility function
-ex ex::to_polynomial(lst & repl_lst) const
-{
-       // Convert lst to exmap
-       exmap m;
-       for (lst::const_iterator it = repl_lst.begin(); it != repl_lst.end(); ++it)
-               m.insert(std::make_pair(it->op(0), it->op(1)));
-
-       ex ret = bp->to_polynomial(m);
-
-       // Convert exmap back to lst
-       repl_lst.remove_all();
-       for (exmap::const_iterator it = m.begin(); it != m.end(); ++it)
-               repl_lst.append(it->first == it->second);
-
-       return ret;
-}
-
 /** Default implementation of ex::to_rational(). This replaces the object with
  *  a temporary symbol. */
 ex basic::to_rational(exmap & repl) const
@@ -2545,7 +2475,7 @@ ex numeric::to_polynomial(exmap & repl) const
 ex power::to_rational(exmap & repl) const
 {
        if (exponent.info(info_flags::integer))
-               return power(basis.to_rational(repl), exponent);
+               return pow(basis.to_rational(repl), exponent);
        else
                return replace_with_symbol(*this, repl);
 }
@@ -2555,17 +2485,17 @@ ex power::to_rational(exmap & repl) const
 ex power::to_polynomial(exmap & repl) const
 {
        if (exponent.info(info_flags::posint))
-               return power(basis.to_rational(repl), exponent);
+               return pow(basis.to_rational(repl), exponent);
        else if (exponent.info(info_flags::negint))
        {
                ex basis_pref = collect_common_factors(basis);
                if (is_exactly_a<mul>(basis_pref) || is_exactly_a<power>(basis_pref)) {
                        // (A*B)^n will be automagically transformed to A^n*B^n
-                       ex t = power(basis_pref, exponent);
+                       ex t = pow(basis_pref, exponent);
                        return t.to_polynomial(repl);
                }
                else
-                       return power(replace_with_symbol(power(basis, _ex_1), repl), -exponent);
+                       return pow(replace_with_symbol(pow(basis, _ex_1), repl), -exponent);
        } 
        else
                return replace_with_symbol(*this, repl);
@@ -2577,17 +2507,15 @@ ex expairseq::to_rational(exmap & repl) const
 {
        epvector s;
        s.reserve(seq.size());
-       epvector::const_iterator i = seq.begin(), end = seq.end();
-       while (i != end) {
-               s.push_back(split_ex_to_pair(recombine_pair_to_ex(*i).to_rational(repl)));
-               ++i;
-       }
+       for (auto & it : seq)
+               s.push_back(split_ex_to_pair(recombine_pair_to_ex(it).to_rational(repl)));
+
        ex oc = overall_coeff.to_rational(repl);
        if (oc.info(info_flags::numeric))
-               return thisexpairseq(s, overall_coeff);
+               return thisexpairseq(std::move(s), overall_coeff);
        else
-               s.push_back(combine_ex_with_coeff_to_pair(oc, _ex1));
-       return thisexpairseq(s, default_overall_coeff());
+               s.push_back(expair(oc, _ex1));
+       return thisexpairseq(std::move(s), default_overall_coeff());
 }
 
 /** Implementation of ex::to_polynomial() for expairseqs. */
@@ -2595,17 +2523,15 @@ ex expairseq::to_polynomial(exmap & repl) const
 {
        epvector s;
        s.reserve(seq.size());
-       epvector::const_iterator i = seq.begin(), end = seq.end();
-       while (i != end) {
-               s.push_back(split_ex_to_pair(recombine_pair_to_ex(*i).to_polynomial(repl)));
-               ++i;
-       }
+       for (auto & it : seq)
+               s.push_back(split_ex_to_pair(recombine_pair_to_ex(it).to_polynomial(repl)));
+
        ex oc = overall_coeff.to_polynomial(repl);
        if (oc.info(info_flags::numeric))
-               return thisexpairseq(s, overall_coeff);
+               return thisexpairseq(std::move(s), overall_coeff);
        else
-               s.push_back(combine_ex_with_coeff_to_pair(oc, _ex1));
-       return thisexpairseq(s, default_overall_coeff());
+               s.push_back(expair(oc, _ex1));
+       return thisexpairseq(std::move(s), default_overall_coeff());
 }
 
 
@@ -2660,7 +2586,7 @@ static ex find_common_factor(const ex & e, ex & factor, exmap & repl)
                                                        else
                                                                v.push_back(t.op(k));
                                                }
-                                               t = (new mul(v))->setflag(status_flags::dynallocated);
+                                               t = dynallocate<mul>(v);
                                                goto term_done;
                                        }
                                }
@@ -2670,7 +2596,7 @@ static ex find_common_factor(const ex & e, ex & factor, exmap & repl)
                        t = x;
 term_done:     ;
                }
-               return (new add(terms))->setflag(status_flags::dynallocated);
+               return dynallocate<add>(terms);
 
        } else if (is_exactly_a<mul>(e)) {
 
@@ -2680,7 +2606,7 @@ term_done:        ;
                for (size_t i=0; i<num; i++)
                        v.push_back(find_common_factor(e.op(i), factor, repl));
 
-               return (new mul(v))->setflag(status_flags::dynallocated);
+               return dynallocate<mul>(v);
 
        } else if (is_exactly_a<power>(e)) {
                const ex e_exp(e.op(1));
@@ -2688,8 +2614,8 @@ term_done:        ;
                        ex eb = e.op(0).to_polynomial(repl);
                        ex factor_local(_ex1);
                        ex pre_res = find_common_factor(eb, factor_local, repl);
-                       factor *= power(factor_local, e_exp);
-                       return power(pre_res, e_exp);
+                       factor *= pow(factor_local, e_exp);
+                       return pow(pre_res, e_exp);
                        
                } else
                        return e.to_polynomial(repl);