lean2/src/kernel/replace_fn.h
Leonardo de Moura 2aa691ccb3 fix(kernel/replace_fn): ignore the cached type in constants
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2013-12-24 21:11:25 -08:00

216 lines
6.8 KiB
C++

/*
Copyright (c) 2013 Microsoft Corporation. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Author: Leonardo de Moura
*/
#pragma once
#include <tuple>
#include "util/buffer.h"
#include "util/interrupt.h"
#include "kernel/expr.h"
#include "kernel/expr_maps.h"
#include "kernel/update_expr.h"
namespace lean {
/**
\brief Default replace_fn postprocessor functional object. It is a
do-nothing object.
*/
class default_replace_postprocessor {
public:
void operator()(expr const &, expr const &) {}
};
/**
\brief Functional for applying <tt>F</tt> to the subexpressions of a given expression.
The signature of \c F is
expr const &, unsigned -> expr
F is invoked for each subexpression \c s of the input expression e.
In a call <tt>F(s, n)</tt>, n is the scope level, i.e., the number of
bindings operators that enclosing \c s.
P is a "post-processing" functional object that is applied to each
pair (old, new)
*/
template<typename F, typename P = default_replace_postprocessor>
class replace_fn {
static_assert(std::is_same<typename std::result_of<F(expr const &, unsigned)>::type, expr>::value,
"replace_fn: return type of F is not expr");
// the return type of P()(e1, e2) should be void
static_assert(std::is_same<typename std::result_of<decltype(std::declval<P>())(expr const &, expr const &)>::type,
void>::value,
"The return type of P()(e1, e2) is not void");
struct frame {
expr m_expr;
unsigned m_offset;
bool m_shared;
unsigned m_index;
frame(expr const & e, unsigned o, bool s):m_expr(e), m_offset(o), m_shared(s), m_index(0) {}
};
typedef buffer<frame> frame_stack;
typedef buffer<expr> result_stack;
expr_cell_offset_map<expr> m_cache;
F m_f;
P m_post;
frame_stack m_fs;
result_stack m_rs;
static bool is_atomic(expr const & e) {
switch (e.kind()) {
case expr_kind::Constant: case expr_kind::Type: case expr_kind::Value:
case expr_kind::Var: case expr_kind::MetaVar:
return true;
default:
return false;
}
}
void save_result(expr const & e, expr const & r, unsigned offset, bool shared) {
if (shared)
m_cache.insert(std::make_pair(expr_cell_offset(e.raw(), offset), r));
m_post(e, r);
m_rs.push_back(r);
}
/**
\brief Visit \c e at the given offset. Return true iff the result is on the
result stack \c m_rs. Return false iff a new frame was pushed on the stack \c m_fs.
The idea is that after the frame is processed, the result will be on the result stack.
*/
bool visit(expr const & e, unsigned offset) {
bool shared = false;
if (is_shared(e)) {
expr_cell_offset p(e.raw(), offset);
auto it = m_cache.find(p);
if (it != m_cache.end()) {
m_rs.push_back(it->second);
return true;
}
shared = true;
}
expr r = m_f(e, offset);
if (is_atomic(r) || !is_eqp(e, r)) {
save_result(e, r, offset, shared);
return true;
} else {
m_fs.emplace_back(e, offset, shared);
return false;
}
}
/**
\brief Return true iff <tt>f.m_index == idx</tt>.
When the result is true, <tt>f.m_index</tt> is incremented.
*/
bool check_index(frame & f, unsigned idx) {
if (f.m_index == idx) {
f.m_index++;
return true;
} else {
return false;
}
}
expr const & rs(int i) {
lean_assert(i < 0);
return m_rs[m_rs.size() + i];
}
void pop_rs(unsigned num) {
m_rs.shrink(m_rs.size() - num);
}
public:
replace_fn(F const & f, P const & p = P()):
m_f(f),
m_post(p) {
}
expr operator()(expr const & e) {
expr r;
visit(e, 0);
while (!m_fs.empty()) {
begin_loop:
check_interrupted();
frame & f = m_fs.back();
expr const & e = f.m_expr;
unsigned offset = f.m_offset;
switch (e.kind()) {
case expr_kind::Constant: case expr_kind::Type: case expr_kind::Value: case expr_kind::Var: case expr_kind::MetaVar:
lean_unreachable(); // LCOV_EXCL_LINE
case expr_kind::App: {
unsigned num = num_args(e);
while (f.m_index < num) {
expr const & c = arg(e, f.m_index);
f.m_index++;
if (!visit(c, offset))
goto begin_loop;
}
r = update_app(e, num, m_rs.end() - num_args(e));
pop_rs(num);
break;
}
case expr_kind::Eq:
if (check_index(f, 0) && !visit(eq_lhs(e), offset))
goto begin_loop;
if (check_index(f, 1) && !visit(eq_rhs(e), offset))
goto begin_loop;
r = update_eq(e, rs(-2), rs(-1));
pop_rs(2);
break;
case expr_kind::Pi: case expr_kind::Lambda:
if (check_index(f, 0) && !visit(abst_domain(e), offset))
goto begin_loop;
if (check_index(f, 1) && !visit(abst_body(e), offset + 1))
goto begin_loop;
r = update_abstraction(e, rs(-2), rs(-1));
pop_rs(2);
break;
case expr_kind::Let:
if (check_index(f, 0) && let_type(e) && !visit(*let_type(e), offset))
goto begin_loop;
if (check_index(f, 1) && !visit(let_value(e), offset))
goto begin_loop;
if (check_index(f, 2) && !visit(let_body(e), offset + 1))
goto begin_loop;
if (let_type(e)) {
r = update_let(e, some_expr(rs(-3)), rs(-2), rs(-1));
pop_rs(3);
} else {
r = update_let(e, none_expr(), rs(-2), rs(-1));
pop_rs(2);
}
break;
}
save_result(e, r, offset, f.m_shared);
m_fs.pop_back();
}
lean_assert(m_rs.size() == 1);
r = m_rs.back();
m_rs.pop_back();
return r;
}
void clear() {
m_cache.clear();
m_fs.clear();
m_rs.clear();
}
};
template<typename F>
expr replace(expr const & e, F f) {
return replace_fn<F>(f)(e);
}
template<typename F, typename P>
expr replace(expr const & e, F f, P p) {
return replace_fn<F, P>(f, p)(e);
}
}