lean2/src/library/simplifier/simp_tactic.cpp

436 lines
17 KiB
C++
Raw Normal View History

/*
Copyright (c) 2015 Microsoft Corporation. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Author: Leonardo de Moura
*/
#include "util/sexpr/option_declarations.h"
#include "kernel/instantiate.h"
#include "library/constants.h"
#include "library/util.h"
#include "library/app_builder.h"
#include "library/relation_manager.h"
#include "library/tactic/expr_to_tactic.h"
#include "library/simplifier/ceqv.h"
#include "library/simplifier/simp_tactic.h"
#include "library/simplifier/simp_rule_set.h"
#ifndef LEAN_DEFAULT_SIMP_SINGLE_PASS
#define LEAN_DEFAULT_SIMP_SINGLE_PASS false
#endif
#ifndef LEAN_DEFAULT_SIMP_BOTTOM_UP
#define LEAN_DEFAULT_SIMP_BOTTOM_UP true
#endif
#ifndef LEAN_DEFAULT_SIMP_BETA_ETA
#define LEAN_DEFAULT_SIMP_BETA_ETA true
#endif
#ifndef LEAN_DEFAULT_SIMP_IOTA
#define LEAN_DEFAULT_SIMP_IOTA true
#endif
#ifndef LEAN_DEFAULT_SIMP_MEMOIZE
#define LEAN_DEFAULT_SIMP_MEMOIZE true
#endif
#ifndef LEAN_DEFAULT_SIMP_MAX_STEPS
#define LEAN_DEFAULT_SIMP_MAX_STEPS 10000
#endif
#ifndef LEAN_DEFAULT_SIMP_TRACE
#define LEAN_DEFAULT_SIMP_TRACE false
#endif
#ifndef LEAN_DEFAULT_SIMP_ASSUMPTIONS
#define LEAN_DEFAULT_SIMP_ASSUMPTIONS false
#endif
#ifndef LEAN_DEFAULT_SIMP_FUNEXT
#define LEAN_DEFAULT_SIMP_FUNEXT true
#endif
#ifndef LEAN_DEFAULT_SIMP_PROPEXT
#define LEAN_DEFAULT_SIMP_PROPEXT true
#endif
namespace lean {
name const * g_simp_single_pass = nullptr;
name const * g_simp_bottom_up = nullptr;
name const * g_simp_beta_eta = nullptr;
name const * g_simp_iota = nullptr;
name const * g_simp_memoize = nullptr;
name const * g_simp_max_steps = nullptr;
name const * g_simp_trace = nullptr;
name const * g_simp_assumptions = nullptr;
name const * g_simp_funext = nullptr;
name const * g_simp_propext = nullptr;
bool get_simp_single_pass(options const & opts) {
return opts.get_bool(*g_simp_single_pass, LEAN_DEFAULT_SIMP_SINGLE_PASS);
}
bool get_simp_bottom_up(options const & opts) {
return opts.get_bool(*g_simp_bottom_up, LEAN_DEFAULT_SIMP_BOTTOM_UP);
}
bool get_simp_beta_eta(options const & opts) {
return opts.get_bool(*g_simp_beta_eta, LEAN_DEFAULT_SIMP_BETA_ETA);
}
bool get_simp_iota(options const & opts) {
return opts.get_bool(*g_simp_iota, LEAN_DEFAULT_SIMP_IOTA);
}
bool get_simp_memoize(options const & opts) {
return opts.get_bool(*g_simp_memoize, LEAN_DEFAULT_SIMP_MEMOIZE);
}
unsigned get_simp_max_steps(options const & opts) {
return opts.get_bool(*g_simp_max_steps, LEAN_DEFAULT_SIMP_MAX_STEPS);
}
bool get_simp_trace(options const & opts) {
return opts.get_bool(*g_simp_trace, LEAN_DEFAULT_SIMP_TRACE);
}
bool get_simp_assumptions(options const & opts) {
return opts.get_bool(*g_simp_assumptions, LEAN_DEFAULT_SIMP_ASSUMPTIONS);
}
bool get_simp_funext(options const & opts) {
return opts.get_bool(*g_simp_funext, LEAN_DEFAULT_SIMP_FUNEXT);
}
bool get_simp_propext(options const & opts) {
return opts.get_bool(*g_simp_propext, LEAN_DEFAULT_SIMP_PROPEXT);
}
expr const * g_simp_tactic = nullptr;
expr mk_simp_tactic_expr(buffer<expr> const & ls, buffer<name> const & ns,
buffer<name> const & ex, optional<expr> const & pre_tac,
location const & loc) {
expr e = mk_expr_list(ls.size(), ls.data());
expr n = ids_to_tactic_expr(ns);
expr x = ids_to_tactic_expr(ex);
expr t;
if (pre_tac) {
t = mk_app(mk_constant(get_option_some_name()), *pre_tac);
} else {
t = mk_constant(get_option_none_name());
}
expr l = mk_location_expr(loc);
expr r = mk_app({*g_simp_tactic, e, n, x, t, l});
return r;
}
class simp_tactic_fn {
2015-07-22 15:39:55 +00:00
public:
enum res_kind { Simplified, Solved, DidNothing };
private:
environment m_env;
io_state m_ios;
name_generator m_ngen;
elaborate_fn m_elab;
optional<tactic> m_tactic;
type_checker m_elab_tc;
app_builder m_app_builder;
// transient state
unsigned m_steps;
goal m_g;
2015-07-22 15:39:55 +00:00
substitution m_subst;
// Remark: the following buffer contains pre-terms that need to be elaborated.
// The 'simp at *' is not very efficient in the current implementation.
// If we have N hypotheses, then m_lemmas_to_process will be processed N+1 times.
buffer<expr> m_lemmas_to_process;
simp_rule_sets m_simp_sets;
// configuration options
bool m_single_pass;
bool m_bottom_up;
bool m_beta_eta;
bool m_iota;
bool m_memoize;
unsigned m_max_steps;
bool m_trace;
bool m_assumptions;
bool m_funext;
bool m_propext;
bool m_standard;
io_state_stream out() const { return regular(m_env, m_ios); }
void set_options(environment const & env, options const & o) {
m_single_pass = get_simp_single_pass(o);
m_bottom_up = get_simp_bottom_up(o);
m_beta_eta = get_simp_beta_eta(o);
m_iota = get_simp_iota(o);
m_memoize = get_simp_memoize(o);
m_max_steps = get_simp_max_steps(o);
m_trace = get_simp_trace(o);
m_assumptions = get_simp_assumptions(o);
if (is_standard(env)) {
m_funext = get_simp_funext(o) && env.find(get_funext_name());
m_propext = get_simp_propext(o) && env.find(get_propext_name());
m_standard = true;
} else {
// TODO(Leo): add support for function extensionality in HoTT mode
m_funext = false;
m_propext = false;
m_standard = false;
}
}
// Add lemmas and assumptions to m_simp_set.
// If hidx is none, then we are elaborating the conclusion, otherwise we are elaborating hypothesis hidx.
// This method destructively updates m_simp_set
void elaborate_lemmas(optional<unsigned> hidx) {
name user("user");
for (expr const & l : m_lemmas_to_process) {
try {
expr new_l; constraints cs;
bool report_unassigned = true;
std::tie(new_l, m_subst, cs) = m_elab(m_g, m_ios.get_options(), m_ngen.mk_child(), l, none_expr(), m_subst, report_unassigned);
if (cs)
throw tactic_exception("invalid 'simp' tactic, fail to resolve generated constraints");
expr new_e = head_beta_reduce(m_elab_tc.infer(new_l).first);
m_simp_sets = add(m_elab_tc, m_simp_sets, user, new_e, new_l);
} catch (exception &) {
if (!hidx) {
// processing conclusion, then report the error
throw;
}
}
}
if (m_assumptions) {
name assump("assumption");
buffer<expr> hyps;
m_g.get_hyps(hyps);
unsigned end = hidx ? *hidx : hyps.size();
for (unsigned i = 0; i < end; i++) {
expr H = hyps[i];
expr H_type = mlocal_type(H);
expr rel, lhs, rhs;
if (is_simp_relation(m_env, H_type, rel, lhs, rhs)) {
// TODO(Leo): we are currently flipping equations when lhs < rhs.
// Should we remove this automatic flipping?
if (get_weight(lhs) >= get_weight(rhs)) {
m_simp_sets = add(m_elab_tc, m_simp_sets, assump, H_type, H);
} else {
// lhs is "smaller" than rhs
// so we try to apply symmetry if available
if (!is_constant(rel))
continue;
name op = const_name(rel);
auto rel_info = get_relation_info(m_env, op);
auto info = get_symm_extra_info(m_env, op);
if (!info || !rel_info)
continue; // relation is not symmetric
buffer<expr> args;
get_app_args(H_type, args);
expr tmp = args[rel_info->get_lhs_pos()];
args[rel_info->get_lhs_pos()] = args[rel_info->get_rhs_pos()];
args[rel_info->get_rhs_pos()] = tmp;
H_type = mk_app(rel, args);
if (auto new_H = m_app_builder.mk_app(info->m_name, H)) {
H = *new_H;
m_simp_sets = add(m_elab_tc, m_simp_sets, assump, H_type, H);
}
}
}
}
}
}
res_kind simp_conclusion() {
elaborate_lemmas(optional<unsigned>());
if (m_trace) {
out() << m_simp_sets;
}
2015-07-22 15:39:55 +00:00
// TODO(Leo)
return DidNothing;
}
bool simp_hyp(unsigned hidx) {
flet<simp_rule_sets> save(m_simp_sets, m_simp_sets);
elaborate_lemmas(optional<unsigned>(hidx));
2015-07-22 15:39:55 +00:00
// TODO(Leo)
return false;
}
// Initialize m_simp_set with information that is context independent
void init_simp_set(buffer<name> const & ns, buffer<name> const & ex) {
// Remark: we cannot initialize explicitly provided lemmas here
// since some of them may depend on hypotheses.
m_simp_sets = get_simp_rule_sets(m_env);
for (name const & n : ns) {
simp_rule_sets tmp_simp_set = get_simp_rule_sets(m_env, n);
m_simp_sets = join(m_simp_sets, tmp_simp_set);
}
m_simp_sets.erase_simp(ex);
}
public:
simp_tactic_fn(environment const & env, io_state const & ios, name_generator && ngen, elaborate_fn const & elab,
buffer<expr> const & ls, buffer<name> const & ns, buffer<name> const & ex,
optional<tactic> const & tac):
m_env(env), m_ios(ios), m_ngen(ngen), m_elab(elab), m_tactic(tac), m_elab_tc(m_env), m_app_builder(m_elab_tc),
m_lemmas_to_process(ls) {
set_options(env, m_ios.get_options());
init_simp_set(ns, ex);
}
2015-07-22 15:39:55 +00:00
std::tuple<res_kind, goal, substitution> operator()(goal const & g, location const & loc, substitution const & s) {
m_g = g;
2015-07-22 15:39:55 +00:00
m_subst = s;
if (loc.is_goal_only()) {
res_kind k = simp_conclusion();
2015-07-22 15:39:55 +00:00
return std::make_tuple(k, m_g, m_subst);
} else {
buffer<expr> hyps;
m_g.get_hyps(hyps);
bool progress = false;
unsigned hidx = 0;
for (expr const & h : hyps) {
if (loc.includes_hypothesis(local_pp_name(h))) {
if (simp_hyp(hidx))
progress = true;
}
hidx++;
}
if (loc.includes_goal()) {
res_kind k = simp_conclusion();
2015-07-22 15:39:55 +00:00
if (k == DidNothing && progress)
k = Simplified;
return std::make_tuple(k, m_g, m_subst);
} else {
return std::make_tuple(progress ? Simplified : DidNothing, m_g, m_subst);
}
}
}
};
tactic mk_simp_tactic(elaborate_fn const & elab, buffer<expr> const & ls, buffer<name> const & ns,
buffer<name> const & ex, optional<tactic> tac, location const & loc) {
return tactic01([=](environment const & env, io_state const & ios, proof_state const & s) {
goals const & gs = s.get_goals();
if (empty(gs)) {
throw_no_goal_if_enabled(s);
return none_proof_state();
}
goal const & g = head(gs);
name_generator new_ngen = s.get_ngen();
simp_tactic_fn simp(env, ios, new_ngen.mk_child(), elab, ls, ns, ex, tac);
goal new_g; simp_tactic_fn::res_kind k; substitution new_subst = s.get_subst();
std::tie(k, new_g, new_subst) = simp(g, loc, new_subst);
switch (k) {
case simp_tactic_fn::Simplified: {
proof_state new_s(s, cons(new_g, tail(gs)), new_subst, new_ngen);
return some_proof_state(new_s);
}
case simp_tactic_fn::Solved: {
proof_state new_s(s, tail(gs), new_subst, new_ngen);
return some_proof_state(new_s);
}
case simp_tactic_fn::DidNothing:
return none_proof_state();
}
lean_unreachable();
});
}
void initialize_simp_tactic() {
name simp_name{"tactic", "simp_tac"};
g_simp_tactic = new expr(mk_constant(simp_name));
register_tac(simp_name,
[](type_checker & tc, elaborate_fn const & elab, expr const & e, pos_info_provider const * p) {
buffer<expr> args;
get_app_args(e, args);
if (args.size() != 5)
throw expr_to_tactic_exception(e, "invalid 'simp' tactic, incorrect number of arguments");
buffer<expr> lemmas;
get_tactic_expr_list_elements(args[0], lemmas, "invalid 'simp' tactic, invalid argument #1");
buffer<name> ns, ex;
get_tactic_id_list_elements(args[1], ns, "invalid 'simp' tactic, invalid argument #2");
get_tactic_id_list_elements(args[2], ex, "invalid 'simp' tactic, invalid argument #3");
optional<tactic> tac;
expr A, t;
if (is_some(args[3], A, t)) {
tac = expr_to_tactic(tc, elab, t, p);
} else if (is_none(args[3], A)) {
// do nothing
} else {
throw expr_to_tactic_exception(e, "invalid 'simp' tactic, invalid argument #4");
}
check_tactic_expr(args[4], "invalid 'simp' tactic, invalid argument #5");
expr loc_expr = get_tactic_expr_expr(args[4]);
if (!is_location_expr(loc_expr))
throw expr_to_tactic_exception(e, "invalid 'simp' tactic, invalid argument #5");
location loc = get_location_expr_location(loc_expr);
return mk_simp_tactic(elab, lemmas, ns, ex, tac, loc);
});
g_simp_single_pass = new name{"simp", "single_pass"};
register_bool_option(*g_simp_single_pass, LEAN_DEFAULT_SIMP_SINGLE_PASS,
"(simp tactic) if false then the simplifier keeps applying simplifications as long as possible");
g_simp_bottom_up = new name{"simp", "bottom_up"};
register_bool_option(*g_simp_bottom_up, LEAN_DEFAULT_SIMP_BOTTOM_UP,
"(simp tactic) if true the simplifier uses a bottom up rewriting strategy, otherwise it uses top down");
g_simp_beta_eta = new name{"simp", "beta_eta"};
register_bool_option(*g_simp_beta_eta, LEAN_DEFAULT_SIMP_BETA_ETA,
"(simp tactic) if true the simplifier applies beta and eta reduction");
g_simp_iota = new name{"simp", "iota"};
register_bool_option(*g_simp_iota, LEAN_DEFAULT_SIMP_IOTA,
"(simp tactic) if true the simplifier applies iota reduction");
g_simp_memoize = new name{"simp", "memoize"};
register_bool_option(*g_simp_memoize, LEAN_DEFAULT_SIMP_MEMOIZE,
"(simp tactic) if true the simplifier caches intermediate results");
g_simp_max_steps = new name{"simp", "max_steps"};
register_unsigned_option(*g_simp_max_steps, LEAN_DEFAULT_SIMP_MAX_STEPS,
"(simp tactic) maximum number of steps that can be performed by the simplifier");
g_simp_trace = new name{"simp", "trace"};
register_bool_option(*g_simp_trace, LEAN_DEFAULT_SIMP_TRACE,
"(simp tactic) if true the simplifier produces an execution trace for debugging purposes");
g_simp_assumptions = new name{"simp", "assumptions"};
register_bool_option(*g_simp_assumptions, LEAN_DEFAULT_SIMP_ASSUMPTIONS,
"(simp tactic) if true assumptions/hypotheses are automatically used as rewriting rules");
g_simp_funext = new name{"simp", "funext"};
register_bool_option(*g_simp_funext, LEAN_DEFAULT_SIMP_FUNEXT,
"(simp tactic) avoid function extensionality even if theorem/axiom is in the environment");
g_simp_propext = new name{"simp", "propext"};
register_bool_option(*g_simp_funext, LEAN_DEFAULT_SIMP_PROPEXT,
"(simp tactic) avoid proposition extensionality even if axiom is in the environment, this option is ignored in HoTT mode");
}
void finalize_simp_tactic() {
delete g_simp_tactic;
delete g_simp_single_pass;
delete g_simp_bottom_up;
delete g_simp_beta_eta;
delete g_simp_iota;
delete g_simp_memoize;
delete g_simp_max_steps;
delete g_simp_trace;
delete g_simp_assumptions;
delete g_simp_funext;
delete g_simp_propext;
}
}