lean2/src/kernel/expr.cpp

654 lines
22 KiB
C++
Raw Normal View History

/*
Copyright (c) 2013-2014 Microsoft Corporation. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Author: Leonardo de Moura
2013-08-01 22:42:06 +00:00
Soonho Kong
*/
#include <vector>
2013-08-01 22:42:06 +00:00
#include <sstream>
#include <string>
#include <algorithm>
#include <limits>
#include "util/list_fn.h"
#include "util/hash.h"
#include "util/buffer.h"
#include "util/object_serializer.h"
#include "util/lru_cache.h"
#include "util/memory_pool.h"
#include "kernel/expr.h"
#include "kernel/expr_eq_fn.h"
#include "kernel/free_vars.h"
#include "kernel/for_each_fn.h"
#ifndef LEAN_INITIAL_EXPR_CACHE_CAPACITY
#define LEAN_INITIAL_EXPR_CACHE_CAPACITY 1024*16
#endif
namespace lean {
unsigned add_weight(unsigned w1, unsigned w2) {
unsigned r = w1 + w2;
if (r < w1)
r = std::numeric_limits<unsigned>::max(); // overflow
return r;
}
unsigned inc_weight(unsigned w) {
if (w < std::numeric_limits<unsigned>::max())
return w+1;
else
return w;
}
static expr * g_dummy = nullptr;
expr::expr():expr(*g_dummy) {}
unsigned hash_levels(levels const & ls) {
unsigned r = 23;
for (auto const & l : ls)
r = hash(hash(l), r);
return r;
}
LEAN_THREAD_VALUE(unsigned, g_hash_alloc_counter, 0);
expr_cell::expr_cell(expr_kind k, unsigned h, bool has_expr_mv, bool has_univ_mv, bool has_local, bool has_param_univ):
m_flags(0),
m_kind(static_cast<unsigned>(k)),
m_has_expr_mv(has_expr_mv),
m_has_univ_mv(has_univ_mv),
m_has_local(has_local),
m_has_param_univ(has_param_univ),
m_hash(h),
m_tag(nulltag),
m_rc(0) {
// m_hash_alloc does not need to be a unique identifier.
// We want diverse hash codes because given expr_cell * c1 and expr_cell * c2,
// if c1 != c2, then there is high probability c1->m_hash_alloc != c2->m_hash_alloc.
// Remark: using pointer address as a hash code is not a good idea.
// - each execution run will behave differently.
// - the hash is not diverse enough
m_hash_alloc = g_hash_alloc_counter;
g_hash_alloc_counter++;
}
void expr_cell::dec_ref(expr & e, buffer<expr_cell*> & todelete) {
if (e.m_ptr) {
expr_cell * c = e.steal_ptr();
lean_assert(!(e.m_ptr));
if (c->dec_ref_core())
todelete.push_back(c);
}
}
optional<bool> expr_cell::is_arrow() const {
// it is stored in bits 0-1
unsigned r = (m_flags & (1+2));
if (r == 0) {
return optional<bool>();
} else if (r == 1) {
return optional<bool>(true);
} else {
lean_assert(r == 2);
return optional<bool>(false);
}
}
void expr_cell::set_is_arrow(bool flag) {
unsigned mask = flag ? 1 : 2;
m_flags |= mask;
lean_assert(is_arrow() && *is_arrow() == flag);
}
void expr_cell::set_tag(tag t) {
m_tag = t;
}
bool is_meta(expr const & e) {
return is_metavar(get_app_fn(e));
}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
// Expr variables
DEF_THREAD_MEMORY_POOL(get_var_allocator, sizeof(expr_var));
expr_var::expr_var(unsigned idx):
expr_cell(expr_kind::Var, idx, false, false, false, false),
m_vidx(idx) {
if (idx == std::numeric_limits<unsigned>::max())
throw exception("invalid free variable index, de Bruijn index is too big");
}
void expr_var::dealloc() {
this->~expr_var();
get_var_allocator().recycle(this);
}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
// Expr constants
DEF_THREAD_MEMORY_POOL(get_const_allocator, sizeof(expr_const));
expr_const::expr_const(name const & n, levels const & ls):
expr_cell(expr_kind::Constant, ::lean::hash(n.hash(), hash_levels(ls)), false, has_meta(ls), false, has_param(ls)),
m_name(n),
m_levels(ls) {
}
void expr_const::dealloc() {
this->~expr_const();
get_const_allocator().recycle(this);
}
// Expr metavariables and local variables
DEF_THREAD_MEMORY_POOL(get_mlocal_allocator, sizeof(expr_mlocal));
expr_mlocal::expr_mlocal(bool is_meta, name const & n, expr const & t):
expr_composite(is_meta ? expr_kind::Meta : expr_kind::Local, n.hash(), is_meta || t.has_expr_metavar(), t.has_univ_metavar(),
!is_meta || t.has_local(), t.has_param_univ(),
1, get_free_var_range(t)),
m_name(n),
m_type(t) {}
void expr_mlocal::dealloc(buffer<expr_cell*> & todelete) {
dec_ref(m_type, todelete);
this->~expr_mlocal();
get_mlocal_allocator().recycle(this);
}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
DEF_THREAD_MEMORY_POOL(get_local_allocator, sizeof(expr_local));
expr_local::expr_local(name const & n, name const & pp_name, expr const & t, binder_info const & bi):
expr_mlocal(false, n, t),
m_pp_name(pp_name),
m_bi(bi) {}
void expr_local::dealloc(buffer<expr_cell*> & todelete) {
dec_ref(m_type, todelete);
this->~expr_local();
get_local_allocator().recycle(this);
}
// Composite expressions
expr_composite::expr_composite(expr_kind k, unsigned h, bool has_expr_mv, bool has_univ_mv,
bool has_local, bool has_param_univ, unsigned w, unsigned fv_range):
expr_cell(k, h, has_expr_mv, has_univ_mv, has_local, has_param_univ),
m_weight(w),
m_free_var_range(fv_range) {}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
// Expr applications
DEF_THREAD_MEMORY_POOL(get_app_allocator, sizeof(expr_app));
expr_app::expr_app(expr const & fn, expr const & arg):
expr_composite(expr_kind::App, ::lean::hash(fn.hash(), arg.hash()),
fn.has_expr_metavar() || arg.has_expr_metavar(),
fn.has_univ_metavar() || arg.has_univ_metavar(),
fn.has_local() || arg.has_local(),
fn.has_param_univ() || arg.has_param_univ(),
inc_weight(add_weight(get_weight(fn), get_weight(arg))),
std::max(get_free_var_range(fn), get_free_var_range(arg))),
m_fn(fn), m_arg(arg) {
m_hash = ::lean::hash(m_hash, m_weight);
}
void expr_app::dealloc(buffer<expr_cell*> & todelete) {
dec_ref(m_fn, todelete);
dec_ref(m_arg, todelete);
this->~expr_app();
get_app_allocator().recycle(this);
}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
static unsigned dec(unsigned k) { return k == 0 ? 0 : k - 1; }
bool operator==(binder_info const & i1, binder_info const & i2) {
return
i1.is_implicit() == i2.is_implicit() &&
i1.is_cast() == i2.is_cast() &&
i1.is_contextual() == i2.is_contextual() &&
i1.is_strict_implicit() == i2.is_strict_implicit();
}
// Expr binders (Lambda, Pi)
DEF_THREAD_MEMORY_POOL(get_binding_allocator, sizeof(expr_binding));
expr_binding::expr_binding(expr_kind k, name const & n, expr const & t, expr const & b, binder_info const & i):
expr_composite(k, ::lean::hash(t.hash(), b.hash()),
t.has_expr_metavar() || b.has_expr_metavar(),
t.has_univ_metavar() || b.has_univ_metavar(),
t.has_local() || b.has_local(),
t.has_param_univ() || b.has_param_univ(),
inc_weight(add_weight(get_weight(t), get_weight(b))),
std::max(get_free_var_range(t), dec(get_free_var_range(b)))),
m_binder(n, t, i),
m_body(b) {
m_hash = ::lean::hash(m_hash, m_weight);
lean_assert(k == expr_kind::Lambda || k == expr_kind::Pi);
}
void expr_binding::dealloc(buffer<expr_cell*> & todelete) {
dec_ref(m_body, todelete);
dec_ref(m_binder.m_type, todelete);
this->~expr_binding();
get_binding_allocator().recycle(this);
}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
// Expr Sort
DEF_THREAD_MEMORY_POOL(get_sort_allocator, sizeof(expr_sort));
expr_sort::expr_sort(level const & l):
expr_cell(expr_kind::Sort, ::lean::hash(l), false, has_meta(l), false, has_param(l)),
m_level(l) {
}
expr_sort::~expr_sort() {}
void expr_sort::dealloc() {
this->~expr_sort();
get_sort_allocator().recycle(this);
}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
// Macro definition
bool macro_definition_cell::lt(macro_definition_cell const &) const { return false; }
bool macro_definition_cell::operator==(macro_definition_cell const & other) const { return typeid(*this) == typeid(other); }
unsigned macro_definition_cell::trust_level() const { return 0; }
format macro_definition_cell::pp(formatter const &) const { return format(get_name()); }
void macro_definition_cell::display(std::ostream & out) const { out << get_name(); }
bool macro_definition_cell::is_atomic_pp(bool, bool) const { return true; }
unsigned macro_definition_cell::hash() const { return get_name().hash(); }
macro_definition::macro_definition(macro_definition_cell * ptr):m_ptr(ptr) { lean_assert(m_ptr); m_ptr->inc_ref(); }
macro_definition::macro_definition(macro_definition const & s):m_ptr(s.m_ptr) { if (m_ptr) m_ptr->inc_ref(); }
macro_definition::macro_definition(macro_definition && s):m_ptr(s.m_ptr) { s.m_ptr = nullptr; }
macro_definition::~macro_definition() { if (m_ptr) m_ptr->dec_ref(); }
macro_definition & macro_definition::operator=(macro_definition const & s) { LEAN_COPY_REF(s); }
macro_definition & macro_definition::operator=(macro_definition && s) { LEAN_MOVE_REF(s); }
bool macro_definition::operator<(macro_definition const & other) const {
if (get_name() == other.get_name())
return m_ptr->lt(*other.m_ptr);
else
return get_name() < other.get_name();
}
static unsigned add_weight(unsigned num, expr const * args) {
unsigned r = 0;
for (unsigned i = 0; i < num; i++)
r = add_weight(r, get_weight(args[i]));
return r;
}
static unsigned get_free_var_range(unsigned num, expr const * args) {
unsigned r = 0;
for (unsigned i = 0; i < num; i++) {
unsigned d = get_free_var_range(args[i]);
if (d > r)
r = d;
}
return r;
}
expr_macro::expr_macro(macro_definition const & m, unsigned num, expr const * args):
expr_composite(expr_kind::Macro,
lean::hash(num, [&](unsigned i) { return args[i].hash(); }, m.hash()),
std::any_of(args, args+num, [](expr const & e) { return e.has_expr_metavar(); }),
std::any_of(args, args+num, [](expr const & e) { return e.has_univ_metavar(); }),
std::any_of(args, args+num, [](expr const & e) { return e.has_local(); }),
std::any_of(args, args+num, [](expr const & e) { return e.has_param_univ(); }),
inc_weight(add_weight(num, args)),
get_free_var_range(num, args)),
m_definition(m),
m_num_args(num) {
m_args = new expr[num];
for (unsigned i = 0; i < m_num_args; i++)
m_args[i] = args[i];
}
void expr_macro::dealloc(buffer<expr_cell*> & todelete) {
for (unsigned i = 0; i < m_num_args; i++) dec_ref(m_args[i], todelete);
delete(this);
}
expr_macro::~expr_macro() {
delete[] m_args;
}
refactor(kernel): add heterogeneous equality back to expr The main motivation is that we will be able to move equalities between universes. For example, suppose we have A : (Type i) B : (Type i) H : @eq (Type j) A B where j > i We didn't find any trick for deducing (@eq (Type i) A B) from H. Before this commit, heterogeneous equality as a constant with type heq : {A B : (Type U)} : A -> B -> Bool So, from H, we would only be able to deduce (@heq (Type j) (Type j) A B) Not being able to move the equality back to a smaller universe is problematic in several cases. I list some instances in the end of the commit message. With this commit, Heterogeneous equality is a special kind of expression. It is not a constant anymore. From H, we can deduce H1 : A == B That is, we are essentially "erasing" the universes when we move to heterogeneous equality. Now, since A and B have (Type i), we can deduce (@eq (Type i) A B) from H1. The proof term is (to_eq (Type i) A B (to_heq (Type j) A B H)) : (@eq (Type i) A B) So, it remains to explain why we need this feature. For example, suppose we want to state the Pi extensionality axiom. axiom hpiext {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)} : A = A' → (∀ x x', x == x' → B x == B' x') → (∀ x, B x) == (∀ x, B' x) This axiom produces an "inflated" equality at (Type U) when we treat heterogeneous equality as a constant. The conclusion (∀ x, B x) == (∀ x, B' x) is syntax sugar for (@heq (Type U) (Type U) (∀ x : A, B x) (∀ x : A', B' x)) Even if A, A', B, B' live in a much smaller universe. As I described above, it doesn't seem to be a way to move this equality back to a smaller universe. So, if we wanted to keep the heterogeneous equality as a constant, it seems we would have to support axiom schemas. That is, hpiext would be parametrized by the universes where A, A', B and B'. Another possibility would be to have universe polymorphism like Agda. None of the solutions seem attractive. So, we decided to have heterogeneous equality as a special kind of expression. And use the trick above to move equalities back to the right universe. BTW, the parser is not creating the new heterogeneous equalities yet. Moreover, kernel.lean still contains a constant name heq2 that is the heterogeneous equality as a constant. Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
2014-02-07 18:07:08 +00:00
// =======================================
// Constructors
#ifdef LEAN_CACHE_EXPRS
typedef lru_cache<expr, expr_hash, is_bi_equal_proc> expr_cache;
LEAN_THREAD_VALUE(bool, g_expr_cache_enabled, true);
MK_THREAD_LOCAL_GET(expr_cache, get_expr_cache, LEAN_INITIAL_EXPR_CACHE_CAPACITY);
bool enable_expr_caching(bool f) {
bool r = g_expr_cache_enabled;
g_expr_cache_enabled = f;
return r;
}
inline expr cache(expr const & e) {
if (g_expr_cache_enabled) {
if (auto r = get_expr_cache().insert(e))
return *r;
}
return e;
}
#else
inline expr cache(expr && e) { return e; }
bool enable_expr_caching(bool) { return true; } // NOLINT
#endif
expr mk_var(unsigned idx) { return cache(expr(new (get_var_allocator().allocate()) expr_var(idx))); }
expr mk_constant(name const & n, levels const & ls) { return cache(expr(new (get_const_allocator().allocate()) expr_const(n, ls))); }
expr mk_macro(macro_definition const & m, unsigned num, expr const * args) { return cache(expr(new expr_macro(m, num, args))); }
expr mk_metavar(name const & n, expr const & t) { return cache(expr(new (get_mlocal_allocator().allocate()) expr_mlocal(true, n, t))); }
expr mk_local(name const & n, name const & pp_n, expr const & t, binder_info const & bi) {
return cache(expr(new (get_local_allocator().allocate()) expr_local(n, pp_n, t, bi)));
}
expr mk_app(expr const & f, expr const & a) {
return cache(expr(new (get_app_allocator().allocate()) expr_app(f, a)));
}
expr mk_binding(expr_kind k, name const & n, expr const & t, expr const & e, binder_info const & i) {
return cache(expr(new (get_binding_allocator().allocate()) expr_binding(k, n, t, e, i)));
}
expr mk_sort(level const & l) { return cache(expr(new (get_sort_allocator().allocate()) expr_sort(l))); }
// =======================================
typedef buffer<expr_cell*> del_buffer;
void expr_cell::dealloc() {
try {
del_buffer todo;
todo.push_back(this);
while (!todo.empty()) {
expr_cell * it = todo.back();
todo.pop_back();
lean_assert(it->get_rc() == 0);
switch (it->kind()) {
case expr_kind::Var: static_cast<expr_var*>(it)->dealloc(); break;
case expr_kind::Macro: static_cast<expr_macro*>(it)->dealloc(todo); break;
case expr_kind::Meta: static_cast<expr_mlocal*>(it)->dealloc(todo); break;
case expr_kind::Local: static_cast<expr_local*>(it)->dealloc(todo); break;
case expr_kind::Constant: static_cast<expr_const*>(it)->dealloc(); break;
case expr_kind::Sort: static_cast<expr_sort*>(it)->dealloc(); break;
case expr_kind::App: static_cast<expr_app*>(it)->dealloc(todo); break;
case expr_kind::Lambda:
case expr_kind::Pi: static_cast<expr_binding*>(it)->dealloc(todo); break;
}
}
} catch (std::bad_alloc&) {
// We need this catch, because push_back may fail when expanding the buffer.
// In this case, we avoid the crash, and "accept" the memory leak.
}
}
// Auxiliary constructors
expr mk_app(expr const & f, unsigned num_args, expr const * args) {
expr r = f;
for (unsigned i = 0; i < num_args; i++)
r = mk_app(r, args[i]);
return r;
}
expr mk_app(unsigned num_args, expr const * args) {
lean_assert(num_args >= 2);
return mk_app(mk_app(args[0], args[1]), num_args - 2, args+2);
}
expr mk_rev_app(expr const & f, unsigned num_args, expr const * args) {
expr r = f;
unsigned i = num_args;
while (i > 0) {
--i;
r = mk_app(r, args[i]);
}
return r;
}
expr mk_rev_app(unsigned num_args, expr const * args) {
lean_assert(num_args >= 2);
return mk_rev_app(mk_app(args[num_args-1], args[num_args-2]), num_args-2, args);
}
expr mk_app_vars(expr const & f, unsigned n) {
expr r = f;
while (n > 0) {
--n;
r = mk_app(r, Var(n));
}
return r;
}
expr const & get_app_args(expr const & e, buffer<expr> & args) {
unsigned sz = args.size();
expr const * it = &e;
while (is_app(*it)) {
args.push_back(app_arg(*it));
it = &(app_fn(*it));
}
std::reverse(args.begin() + sz, args.end());
return *it;
}
void flat_app(expr const & e, buffer<expr> & args) {
unsigned i = args.size();
args.push_back(expr());
expr const & f = get_app_args(e, args);
args[i] = f;
}
expr const & get_app_rev_args(expr const & e, buffer<expr> & args) {
expr const * it = &e;
while (is_app(*it)) {
args.push_back(app_arg(*it));
it = &(app_fn(*it));
}
return *it;
}
expr const & get_app_fn(expr const & e) {
expr const * it = &e;
while (is_app(*it)) {
it = &(app_fn(*it));
}
return *it;
}
unsigned get_app_num_args(expr const & e) {
expr const * it = &e;
unsigned n = 0;
while (is_app(*it)) {
it = &(app_fn(*it));
n++;
}
return n;
}
static name * g_default_name = nullptr;
static name const & get_default_var_name() {
return *g_default_name;
}
static name const & g_default_var_name = get_default_var_name(); // force it to be initialized
bool is_default_var_name(name const & n) { return n == get_default_var_name(); }
expr mk_arrow(expr const & t, expr const & e) { return mk_pi(get_default_var_name(), t, e); }
expr mk_pi(unsigned sz, expr const * domain, expr const & range) {
expr r = range;
unsigned i = sz;
while (i > 0) {
--i;
r = mk_pi(name(g_default_var_name, i), domain[i], r);
}
return r;
}
static expr * g_Prop = nullptr;
static expr * g_Type1 = nullptr;
expr mk_Prop() { return *g_Prop; }
expr mk_Type() { return *g_Type1; }
unsigned get_weight(expr const & e) {
switch (e.kind()) {
case expr_kind::Var: case expr_kind::Constant: case expr_kind::Sort:
case expr_kind::Meta: case expr_kind::Local:
return 1;
case expr_kind::Lambda: case expr_kind::Pi: case expr_kind::Macro:
case expr_kind::App:
return static_cast<expr_composite*>(e.raw())->m_weight;
}
lean_unreachable(); // LCOV_EXCL_LINE
}
bool operator==(expr const & a, expr const & b) { return expr_eq_fn()(a, b); }
bool is_bi_equal(expr const & a, expr const & b) { return expr_eq_fn(true)(a, b); }
expr copy_tag(expr const & e, expr && new_e) {
tag t = e.get_tag();
if (t != nulltag)
new_e.set_tag(t);
return new_e;
}
expr update_app(expr const & e, expr const & new_fn, expr const & new_arg) {
if (!is_eqp(app_fn(e), new_fn) || !is_eqp(app_arg(e), new_arg))
return copy_tag(e, mk_app(new_fn, new_arg));
else
return e;
}
expr update_rev_app(expr const & e, unsigned num, expr const * new_args) {
expr const * it = &e;
for (unsigned i = 0; i < num - 1; i++) {
if (!is_app(*it) || !is_eqp(app_arg(*it), new_args[i]))
return copy_tag(e, mk_rev_app(num, new_args));
it = &app_fn(*it);
}
if (!is_eqp(*it, new_args[num - 1]))
return copy_tag(e, mk_rev_app(num, new_args));
return e;
}
expr update_binding(expr const & e, expr const & new_domain, expr const & new_body) {
if (!is_eqp(binding_domain(e), new_domain) || !is_eqp(binding_body(e), new_body))
return copy_tag(e, mk_binding(e.kind(), binding_name(e), new_domain, new_body, binding_info(e)));
else
return e;
}
expr update_binding(expr const & e, expr const & new_domain, expr const & new_body, binder_info const & bi) {
if (!is_eqp(binding_domain(e), new_domain) || !is_eqp(binding_body(e), new_body) || bi != binding_info(e))
return copy_tag(e, mk_binding(e.kind(), binding_name(e), new_domain, new_body, bi));
else
return e;
}
expr update_mlocal(expr const & e, expr const & new_type) {
if (is_eqp(mlocal_type(e), new_type))
return e;
else if (is_metavar(e))
return copy_tag(e, mk_metavar(mlocal_name(e), new_type));
else
return copy_tag(e, mk_local(mlocal_name(e), local_pp_name(e), new_type, local_info(e)));
}
expr update_local(expr const & e, expr const & new_type, binder_info const & bi) {
if (is_eqp(mlocal_type(e), new_type) && local_info(e) == bi)
return e;
else
return copy_tag(e, mk_local(mlocal_name(e), local_pp_name(e), new_type, bi));
}
expr update_local(expr const & e, binder_info const & bi) {
return update_local(e, mlocal_type(e), bi);
}
expr update_sort(expr const & e, level const & new_level) {
if (!is_eqp(sort_level(e), new_level))
return copy_tag(e, mk_sort(new_level));
else
return e;
}
expr update_constant(expr const & e, levels const & new_levels) {
if (!is_eqp(const_levels(e), new_levels))
return copy_tag(e, mk_constant(const_name(e), new_levels));
else
return e;
}
expr update_macro(expr const & e, unsigned num, expr const * args) {
if (num == macro_num_args(e)) {
unsigned i = 0;
for (i = 0; i < num; i++) {
if (!is_eqp(macro_arg(e, i), args[i]))
break;
}
if (i == num)
return e;
}
return copy_tag(e, mk_macro(to_macro(e)->m_definition, num, args));
}
bool is_atomic(expr const & e) {
switch (e.kind()) {
case expr_kind::Constant: case expr_kind::Sort:
case expr_kind::Var:
return true;
case expr_kind::Macro:
return to_macro(e)->get_num_args() == 0;
case expr_kind::App: case expr_kind::Meta:
case expr_kind::Local: case expr_kind::Lambda:
case expr_kind::Pi:
return false;
}
lean_unreachable(); // LCOV_EXCL_LINE
}
bool is_arrow(expr const & t) {
optional<bool> r = t.raw()->is_arrow();
if (r) {
return *r;
} else {
bool res = is_pi(t) && !has_free_var(binding_body(t), 0);
t.raw()->set_is_arrow(res);
return res;
}
}
bool has_expr_metavar_strict(expr const & e) {
if (!has_expr_metavar(e))
return false;
bool found = false;
for_each(e, [&](expr const & e, unsigned) {
if (found || !has_expr_metavar(e)) return false;
if (is_metavar(e)) { found = true; return false; }
if (is_local(e)) return false; // do not visit type
return true;
});
return found;
}
static bool has_free_var_in_domain(expr const & b, unsigned vidx) {
if (is_pi(b)) {
return has_free_var(binding_domain(b), vidx) || has_free_var_in_domain(binding_body(b), vidx+1);
} else {
return false;
}
}
expr infer_implicit(expr const & t, unsigned num_params, bool strict) {
if (num_params == 0) {
return t;
} else if (is_pi(t)) {
expr new_body = infer_implicit(binding_body(t), num_params-1, strict);
if (binding_info(t).is_implicit() || binding_info(t).is_strict_implicit()) {
// argument is already marked as implicit
return update_binding(t, binding_domain(t), new_body);
} else if ((strict && has_free_var_in_domain(new_body, 0)) ||
(!strict && has_free_var(new_body, 0))) {
return update_binding(t, binding_domain(t), new_body, mk_implicit_binder_info());
} else {
return update_binding(t, binding_domain(t), new_body);
}
} else {
return t;
}
}
expr infer_implicit(expr const & t, bool strict) {
return infer_implicit(t, std::numeric_limits<unsigned>::max(), strict);
}
void initialize_expr() {
g_dummy = new expr(mk_var(0));
g_default_name = new name("a");
g_Type1 = new expr(mk_sort(mk_level_one()));
g_Prop = new expr(mk_sort(mk_level_zero()));
}
void finalize_expr() {
delete g_Prop;
delete g_Type1;
delete g_dummy;
delete g_default_name;
}
}