adapt code generation

This commit is contained in:
Mark Thom
2022-10-17 22:56:08 -06:00
committed by Mark
parent b9c9de5222
commit e41d1b319b
8 changed files with 700 additions and 427 deletions

View File

@@ -60,6 +60,9 @@ pub(crate) trait Allocator {
fn take_bindings(self) -> AllocVarDict;
fn max_reg_allocated(&self) -> usize;
// TODO: wha.. why?? grrr. it drains the VarStatus data from vs (which it owns!)
// into self.bindings and perm_vs after all is computed (i.e. vs.populate_restricting_sets()
// and vs.set_perm_vals(has_deep_cut) have both been called).
fn drain_var_data<'a>(
&mut self,
vs: VariableFixtures<'a>,

View File

@@ -84,7 +84,7 @@ type VariableFixture<'a> = (VarStatus, Vec<&'a Cell<VarReg>>);
#[derive(Debug)]
pub(crate) struct VariableFixtures<'a> {
perm_vars: IndexMap<Var, VariableFixture<'a>>,
last_chunk_temp_vars: IndexSet<Var>,
last_chunk_temp_vars: IndexSet<Var>, // TODO: has no use at all!
}
impl<'a> VariableFixtures<'a> {

View File

@@ -1,6 +1,7 @@
use crate::arena::*;
use crate::atom_table::*;
use crate::instructions::*;
use crate::machine::disjuncts::VarRecord;
use crate::machine::heap::*;
use crate::machine::loader::PredicateQueue;
use crate::machine::machine_errors::*;
@@ -34,7 +35,7 @@ pub type JumpStub = Vec<Term>;
#[derive(Debug, Clone)]
pub enum TopLevel {
Fact(Term), // Term, line_num, col_num
Fact(Fact), // Term, line_num, col_num
Predicate(Predicate),
Query(Vec<QueryTerm>),
Rule(Rule), // Rule, line_num, col_num
@@ -82,10 +83,11 @@ pub enum CallPolicy {
pub enum QueryTerm {
// register, clause type, subterms, clause call policy.
Clause(Cell<RegType>, ClauseType, Vec<Term>, CallPolicy),
BlockedCut, // a cut which is 'blocked by letters', like the P term in P -> Q.
UnblockedCut(Cell<VarReg>),
Cut,
Not(Vec<QueryTerm>),
IfThen(Vec<QueryTerm>, Vec<QueryTerm>),
Branch(Vec<Vec<QueryTerm>>),
GetLevelAndUnify(Cell<VarReg>, Var),
Jump(JumpStub), // SOON: Branch(Vec<QueryTerm>),
}
impl QueryTerm {
@@ -99,17 +101,24 @@ impl QueryTerm {
pub(crate) fn arity(&self) -> usize {
match self {
&QueryTerm::Clause(_, _, ref subterms, ..) => subterms.len(),
&QueryTerm::BlockedCut | &QueryTerm::UnblockedCut(..) => 0,
&QueryTerm::Jump(ref vars) => vars.len(),
&QueryTerm::GetLevelAndUnify(..) => 1,
&QueryTerm::Cut | &QueryTerm::Branch(_) => 0,
&QueryTerm::IfThen(..) => 2,
&QueryTerm::Not(_) | &QueryTerm::GetLevelAndUnify(..) => 1,
}
}
}
#[derive(Debug, Clone)]
pub struct Fact {
pub(crate) head: Term,
pub(crate) var_records: Vec<VarRecord>,
}
#[derive(Debug, Clone)]
pub struct Rule {
pub(crate) head: (Atom, Vec<Term>, QueryTerm),
pub(crate) clauses: Vec<QueryTerm>,
pub(crate) var_records: Vec<VarRecord>,
}
#[derive(Clone, Debug, Hash)]

View File

@@ -27,7 +27,6 @@ pub mod instructions {
include!(concat!(env!("OUT_DIR"), "/instructions.rs"));
}
mod iterators;
mod disjuncts;
pub mod machine;
mod raw_block;
pub mod read;

640
src/machine/disjuncts.rs Normal file
View File

@@ -0,0 +1,640 @@
/*
================================================================================
This is a disjunction compilation experiment attempting to adapt the
paper "Compiling Large Disjunctions" to Scryer Prolog.
================================================================================
*/
use crate::atom_table::*;
use crate::forms::*;
use crate::instructions::*;
use crate::iterators::*;
use crate::machine::loader::*;
use crate::machine::machine_errors::CompilationError;
use crate::machine::preprocessor::*;
use crate::parser::ast::*;
use crate::parser::rug::Rational;
use indexmap::{IndexMap, IndexSet};
use std::cell::Cell;
use std::cmp::Ordering;
use std::hash::{Hash, Hasher};
use std::ops::{Deref, DerefMut};
#[derive(Debug, Clone)]
struct BranchNumber {
branch_num: Rational,
delta: Rational,
}
impl Default for BranchNumber {
fn default() -> Self {
Self {
branch_num: Rational::from(1 << 10),
delta: Rational::from(1),
}
}
}
impl PartialEq<BranchNumber> for BranchNumber {
#[inline]
fn eq(&self, rhs: &BranchNumber) -> bool {
self.branch_num == rhs.branch_num
}
}
impl Eq for BranchNumber {}
impl Hash for BranchNumber {
#[inline(always)]
fn hash<H: Hasher>(&self, hasher: &mut H) {
self.branch_num.hash(hasher)
}
}
impl PartialOrd<BranchNumber> for BranchNumber {
#[inline]
fn partial_cmp(&self, rhs: &BranchNumber) -> Option<Ordering> {
self.branch_num.partial_cmp(&rhs.branch_num)
}
}
impl BranchNumber {
fn split(&self) -> BranchNumber {
BranchNumber {
branch_num: self.branch_num.clone() + &self.delta / Rational::from(2),
delta: &self.delta / Rational::from(4),
}
}
fn incr_by_delta(&self) -> BranchNumber {
BranchNumber {
branch_num: self.branch_num.clone() + &self.delta,
delta: self.delta.clone(),
}
}
fn halve_delta(&self) -> BranchNumber {
BranchNumber {
branch_num: self.branch_num.clone(),
delta : &self.delta / Rational::from(2),
}
}
}
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct ChunkInfo {
chunk_num: usize,
vars: Vec<VarPtr>, // pointer to incidence
}
impl ChunkInfo {
fn new(chunk_num: usize) -> Self {
ChunkInfo { chunk_num, vars: vec![] }
}
}
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct BranchInfo {
branch_num: BranchNumber,
chunks: Vec<ChunkInfo>,
}
impl BranchInfo {
fn new(branch_num: BranchNumber) -> Self {
Self { branch_num, chunks: vec![] }
}
}
type BranchMapInt = IndexMap<Var, Vec<BranchInfo>>;
#[derive(Debug, Clone)]
pub struct BranchMap(BranchMapInt);
impl Deref for BranchMap {
type Target = BranchMapInt;
#[inline(always)]
fn deref(&self) -> &BranchMapInt {
&self.0
}
}
impl DerefMut for BranchMap {
#[inline(always)]
fn deref_mut(&mut self) -> &mut BranchMapInt {
&mut self.0
}
}
type RootSet = IndexSet<BranchNumber>;
enum TraversalState {
BuildDisjunct(usize), // construct a QueryTerm::Branch with number of disjuncts.
BuildIf(usize, Term), // build the P term of P -> Q
BuildThen(usize, Vec<QueryTerm>), // build the Q term of P -> Q
BuildNot(usize), // build the P term of \+ P
ResetCallPolicy(CallPolicy),
Term(Term),
AddBranchNum(BranchNumber), // set current_branch_number, add it to the root set
RemoveBranchNum, // remove latest branch number from the root set
RepBranchNum(BranchNumber), // replace current_branch_number and the latest in the root set
IncrChunkNum, // increment self.current_chunk_number
}
impl Term {
#[inline]
fn is_var(&self) -> bool {
if let Term::Var(..) = self {
true
} else {
false
}
}
#[inline]
fn is_compound(&self) -> bool {
match self {
Term::Clause(..) | Term::Cons(..) => true,
_ => false,
}
}
}
pub struct VariableClassifier {
call_policy: CallPolicy,
current_branch_num: BranchNumber,
current_chunk_num: usize,
branch_map: BranchMap,
root_set: RootSet,
}
#[derive(Debug)]
pub enum VarClassification {
Void,
Temp,
Perm,
}
pub struct VarRecord {
pub classification: VarClassification,
pub chunk_occurrences: Vec<usize>,
pub num_occurrences: usize,
}
pub type ClassifyFactResult = (Term, Vec<VarRecord>);
pub type ClassifyRuleResult = (Term, Vec<QueryTerm>, Vec<VarRecord>);
fn merge_branch_seq<Iter: Iterator<Item = BranchInfo>>(branches: Iter) -> BranchInfo {
let mut branch_info = BranchInfo::new(BranchNumber::default());
for mut branch in branches {
branch_info.branch_num = branch.branch_num;
if let Some(last_chunk) = branch_info.chunks.last_mut() {
if let Some(first_moved_chunk) = branch.chunks.first_mut() {
if last_chunk.chunk_num == first_moved_chunk.chunk_num {
last_chunk.vars.extend(first_moved_chunk.vars.drain(..));
branch_info.chunks.extend(branch.chunks.drain(1 ..));
continue;
}
}
}
branch_info.chunks.extend(branch.chunks.drain(..));
}
branch_info.branch_num.delta *= 2;
branch_info.branch_num.branch_num -= &branch_info.branch_num.delta;
branch_info
}
impl VariableClassifier {
pub fn new(call_policy: CallPolicy) -> Self {
Self {
call_policy,
current_branch_num: BranchNumber::default(),
current_chunk_num: 0,
branch_map: BranchMap(BranchMapInt::new()),
root_set: RootSet::new(),
}
}
pub fn classify_fact(mut self, term: Term) -> Result<ClassifyFactResult, CompilationError> {
self.classify_head_variables(&term)?;
Ok((term, self.branch_map.separate_and_classify_variables()))
}
pub fn classify_rule<'a, LS: LoadState<'a>>(
mut self,
loader: &mut Loader<'a, LS>,
head: Term,
body: Term,
) -> Result<ClassifyRuleResult, CompilationError> {
self.classify_head_variables(&head)?;
let query_terms = self.classify_body_variables(loader, body)?;
Ok((head, query_terms, self.branch_map.separate_and_classify_variables()))
}
/*
pub fn to_branch_map(mut self, term: Term) -> Result<ClassifierResult, CompilationError> {
self.root_set.insert(BranchNumber::default());
let (head_term, query_terms) = match term {
Term::Clause(_, atom!(":-"), terms) if terms.len() == 2 => {
let head_term = terms[0];
self.classify_head_variables(&head_term)?;
(head_term, self.classify_body_variables(terms[1])?)
}
_ => {
self.classify_head_variables(&term)?;
(term, vec![])
}
};
self.merge_branches();
Ok((head_term, query_terms, self.branch_map))
}
*/
fn merge_branches(&mut self) {
for branches in self.branch_map.values_mut() {
let mut old_branches = std::mem::replace(branches, vec![]);
while let Some(last_branch_num) = old_branches.last().map(|bi| &bi.branch_num) {
let mut old_branches_len = old_branches.len();
for (rev_idx, bi) in old_branches.iter().rev().enumerate() {
if &bi.branch_num > last_branch_num {
old_branches_len = old_branches.len() - rev_idx;
}
}
let iter = old_branches.drain(old_branches_len - 1 ..);
branches.push(merge_branch_seq(iter));
}
branches.reverse();
}
}
fn probe_body_term(&mut self, term: &Term) {
// true to iterate the root, which may be a variable!
for term_ref in breadth_first_iter(term, true) {
if let TermRef::Var(_, _, var_name) = term_ref {
self.probe_body_var(Var::from(var_name));
}
}
}
fn probe_body_var(&mut self, var_name: Var) {
let branch_info_v = self.branch_map.entry(var_name)
.or_insert_with(|| vec![]);
let needs_new_branch = if let Some(last_bi) = branch_info_v.last() {
!self.root_set.contains(&last_bi.branch_num)
} else {
true
};
if needs_new_branch {
branch_info_v.push(BranchInfo::new(self.current_branch_num.clone()));
}
let branch_info = branch_info_v.last_mut().unwrap();
let needs_new_chunk = if let Some(last_ci) = branch_info.chunks.last() {
last_ci.chunk_num != self.current_chunk_num
} else {
true
};
if needs_new_chunk {
branch_info.chunks.push(ChunkInfo::new(self.current_chunk_num));
}
let chunk_info = branch_info.chunks.last_mut().unwrap();
chunk_info.vars.push(VarPtr::from(&var_name));
}
fn classify_head_variables(&mut self, term: &Term) -> Result<(), CompilationError> {
match term {
Term::Clause(..) | Term::Literal(_, Literal::Atom(_)) => {
}
_ => return Err(CompilationError::InvalidRuleHead),
}
// false argument to breadth_first_iter because the root is not iterable.
for term_ref in breadth_first_iter(term, false) {
if let TermRef::Var(_, _, var_name) = term_ref {
// the body of the if let here is an inlined
// "probe_head_var". note the difference between it
// and "probe_body_var".
let branch_info_v = self.branch_map.entry(Var::from(var_name))
.or_insert_with(|| vec![]);
let needs_new_branch = branch_info_v.is_empty();
if needs_new_branch {
branch_info_v.push(BranchInfo::new(self.current_branch_num.clone()));
}
let branch_info = branch_info_v.last_mut().unwrap();
let needs_new_chunk = branch_info.chunks.is_empty();
if needs_new_chunk {
branch_info.chunks.push(ChunkInfo::new(self.current_chunk_num));
}
let chunk_info = branch_info.chunks.last_mut().unwrap();
chunk_info.vars.push(VarPtr::from(&var_name));
}
}
Ok(())
}
fn classify_body_variables<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
term: Term,
) -> Result<Vec<QueryTerm>, CompilationError> {
let mut state_stack = vec![TraversalState::Term(term)];
let mut build_stack = vec![];
while let Some(traversal_st) = state_stack.pop() {
match traversal_st {
TraversalState::AddBranchNum(branch_num) => {
self.root_set.insert(branch_num.clone());
self.current_branch_num = branch_num;
}
TraversalState::RemoveBranchNum => {
self.root_set.pop();
}
TraversalState::RepBranchNum(branch_num) => {
self.root_set.pop();
self.root_set.insert(branch_num.clone());
self.current_branch_num = branch_num;
}
TraversalState::IncrChunkNum => {
self.current_chunk_num += 1;
}
TraversalState::BuildDisjunct(preceding_len) => {
let iter = build_stack.drain(preceding_len ..);
if let QueryTerm::Branch(ref mut disjuncts) = &mut build_stack[preceding_len] {
disjuncts.push(iter.collect());
}
}
TraversalState::BuildIf(preceding_len, then_term) => {
let iter = build_stack.drain(preceding_len ..);
let build_stack_len = build_stack.len();
state_stack.push(TraversalState::BuildThen(build_stack_len, iter.collect()));
}
TraversalState::BuildThen(preceding_len, if_terms) => {
let iter = build_stack.drain(preceding_len ..);
build_stack.push(QueryTerm::IfThen(if_terms, iter.collect()));
}
TraversalState::BuildNot(preceding_len) => {
let iter = build_stack.drain(preceding_len ..);
build_stack.push(QueryTerm::Not(iter.collect()));
}
TraversalState::ResetCallPolicy(call_policy) => {
self.call_policy = call_policy;
}
TraversalState::Term(term) => {
match term {
Term::Clause(_, atom!(","), terms) if terms.len() == 2 => {
state_stack.extend(
unfold_by_str(terms[1], atom!(","))
.into_iter()
.rev()
.map(TraversalState::Term),
);
state_stack.push(TraversalState::Term(terms[0]));
}
Term::Clause(_, atom!(";"), terms) if terms.len() == 2 => {
let first_branch_num = self.current_branch_num.split();
let branches: Vec<_> = std::iter::once(terms[0])
.chain(unfold_by_str(terms[1], atom!(";")).into_iter())
.collect();
let mut branch_numbers = vec![first_branch_num];
for idx in 1 .. branches.len() {
let succ_branch_number = branch_numbers[idx - 1].incr_by_delta();
branch_numbers.push(if idx + 1 < branches.len() {
succ_branch_number.split()
} else {
succ_branch_number
});
}
let build_stack_len = build_stack.len();
build_stack.push(QueryTerm::Branch(vec![]));
state_stack.push(TraversalState::BuildDisjunct(build_stack_len));
state_stack.push(TraversalState::RepBranchNum(
self.current_branch_num.halve_delta(),
));
let iter = branches.into_iter().zip(branch_numbers.into_iter());
for (term, branch_num) in iter.rev() {
state_stack.push(TraversalState::BuildDisjunct(build_stack_len));
state_stack.push(TraversalState::RemoveBranchNum);
state_stack.push(TraversalState::Term(term));
state_stack.push(TraversalState::AddBranchNum(branch_num));
}
}
Term::Clause(_, atom!("->"), mut terms) if terms.len() == 2 => {
let then_term = terms.pop().unwrap();
let if_term = terms.pop().unwrap();
let build_stack_len = build_stack.len();
state_stack.push(TraversalState::BuildIf(build_stack_len, then_term));
state_stack.push(TraversalState::Term(if_term));
}
Term::Clause(_, atom!("\\+"), terms) if terms.len() == 1 => {
let build_stack_len = build_stack.len();
state_stack.push(TraversalState::BuildNot(build_stack_len));
state_stack.push(TraversalState::Term(terms[0]));
}
Term::Clause(_, atom!("$get_level"), terms) if terms.len() == 1 => {
state_stack.push(TraversalState::IncrChunkNum);
if let Term::Var(_, ref var) = &terms[0] {
build_stack.push(QueryTerm::GetLevelAndUnify(Cell::default(), var.clone()));
} else {
return Err(CompilationError::InadmissibleQueryTerm);
}
}
Term::Clause(_, atom!(":"), mut terms) if terms.len() == 2 => {
let predicate_name = terms.pop().unwrap();
let module_name = terms.pop().unwrap();
match (module_name, predicate_name) {
(
Term::Literal(_, Literal::Atom(module_name)),
Term::Literal(_, Literal::Atom(predicate_name)),
) => {
if !ClauseType::is_inbuilt(name, 0) {
state_stack.push(TraversalState::IncrChunkNum);
}
build_stack.push(
qualified_clause_to_query_term(
loader,
module_name,
predicate_name,
vec![],
self.call_policy,
),
);
}
(
Term::Literal(_, Literal::Atom(module_name)),
Term::Clause(_, name, terms),
) => {
if !ClauseType::is_inbuilt(name, terms.len()) {
state_stack.push(TraversalState::IncrChunkNum);
}
build_stack.push(
qualified_clause_to_query_term(
loader,
module_name,
name,
terms,
self.call_policy,
),
);
}
(module_name, predicate_name) => {
state_stack.push(TraversalState::IncrChunkNum);
terms.push(module_name);
terms.push(predicate_name);
build_stack.push(
clause_to_query_term(
loader,
atom!("call"),
vec![Term::Clause(Cell::default(), atom!(":"), terms)],
self.call_policy,
),
);
}
}
}
Term::Clause(cell, atom!("$call_with_inference_counting"), terms) if terms.len() == 2 => {
state_stack.push(TraversalState::ResetCallPolicy(self.call_policy));
state_stack.push(TraversalState::Term(terms[0]));
self.call_policy = CallPolicy::Counted;
}
Term::Clause(cell, name, terms) => {
if !ClauseType::is_inbuilt(name, terms.len()) {
state_stack.push(TraversalState::IncrChunkNum);
}
for term in terms.iter() {
self.probe_body_term(term);
}
build_stack.push(
clause_to_query_term(
loader,
name,
terms,
self.call_policy,
),
);
}
Term::Literal(_, Literal::Atom(atom!("!"))) |
Term::Literal(_, Literal::Char('!')) => {
build_stack.push(QueryTerm::Cut);
}
Term::Literal(cell, Literal::Atom(name)) => {
if !ClauseType::is_inbuilt(name, 0) {
state_stack.push(TraversalState::IncrChunkNum);
}
build_stack.push(
clause_to_query_term(
loader,
name,
vec![],
self.call_policy,
),
);
}
_ => {
return Err(CompilationError::InadmissibleQueryTerm);
}
}
}
}
}
Ok(build_stack)
}
}
impl BranchMap {
pub fn separate_and_classify_variables(&mut self) -> Vec<VarRecord> {
let mut var_num = 0usize;
let mut records = vec![];
for branches in self.values_mut() {
for branch in branches.iter_mut() {
let mut num_occurrences = 0;
let mut chunk_occurrences = vec![];
for chunk in branch.chunks.iter_mut() {
num_occurrences += chunk.vars.len();
for var in chunk.vars.iter_mut() {
var.set(Var::Generated(var_num));
}
chunk_occurrences.push(chunk.chunk_num);
}
let classification = if branch.chunks.len() > 1 {
VarClassification::Perm
} else {
branch.chunks
.first()
.map(|chunk| if chunk.vars.len() > 1 {
VarClassification::Temp
} else {
VarClassification::Void
})
.unwrap_or(VarClassification::Void)
};
records.push(VarRecord { classification, chunk_occurrences, num_occurrences });
var_num += 1;
}
}
debug_assert_eq!(records.len(), var_num);
records
}
}

View File

@@ -441,7 +441,7 @@ impl<'a, LS: LoadState<'a>> Loader<'a, LS> {
term: Term,
preprocessor: &mut Preprocessor,
) -> Result<PredicateClause, SessionError> {
let tl = preprocessor.try_term_to_tl(self, term, CutContext::BlocksCuts)?;
let tl = preprocessor.try_term_to_tl(self, term)?;
Ok(match tl {
TopLevel::Fact(fact) => PredicateClause::Fact(fact),

View File

@@ -16,6 +16,7 @@ pub mod machine_state;
pub mod machine_state_impl;
pub mod mock_wam;
pub mod partial_string;
pub mod disjuncts;
pub mod preprocessor;
pub mod stack;
pub mod streams;

View File

@@ -2,7 +2,7 @@ use crate::atom_table::*;
use crate::codegen::CodeGenSettings;
use crate::forms::*;
use crate::instructions::*;
use crate::iterators::*;
use crate::machine::disjuncts::*;
use crate::machine::loader::*;
use crate::machine::machine_errors::*;
use crate::parser::ast::*;
@@ -13,21 +13,6 @@ use std::cell::Cell;
use std::collections::VecDeque;
use std::convert::TryFrom;
/*
* The preprocessor fabricates if-then-else ( .. -> ... ; ...)
* clauses into nameless standalone predicates, which it queues for
* later preprocessing and compilation. Fabricated predicates inherit
* explicit "cut variables" from the handwritten predicate
* surrounding their source if-then-else. They must be specially
* handled.
*/
#[derive(Clone, Copy, Debug)]
pub(crate) enum CutContext {
BlocksCuts,
HasCutVariable,
}
pub(crate) fn fold_by_str<I>(terms: I, mut term: Term, sym: Atom) -> Term
where
I: DoubleEndedIterator<Item = Term>,
@@ -131,6 +116,13 @@ fn setup_module_export(
})
}
pub(crate) fn build_rule_body(vars: &[Term], body_term: Term) -> Term {
let head_term = Term::Clause(Cell::default(), atom!(""), vars.iter().cloned().collect());
let rule = vec![head_term, body_term];
Term::Clause(Cell::default(), atom!(":-"), rule)
}
pub(super) fn setup_module_export_list(
mut export_list: Term,
atom_tbl: &mut AtomTable,
@@ -324,110 +316,6 @@ fn setup_meta_predicate<'a, LS: LoadState<'a>>(
}
}
fn merge_clauses(tls: &mut VecDeque<TopLevel>) -> Result<TopLevel, CompilationError> {
let mut clauses = vec![];
while let Some(tl) = tls.pop_front() {
match tl {
TopLevel::Query(_) if clauses.is_empty() && tls.is_empty() => {
return Ok(tl);
}
TopLevel::Query(_) => {
return Err(CompilationError::InconsistentEntry);
}
TopLevel::Fact(fact) => {
let clause = PredicateClause::Fact(fact);
clauses.push(clause);
}
TopLevel::Rule(rule) => {
let clause = PredicateClause::Rule(rule);
clauses.push(clause);
}
TopLevel::Predicate(predicate) => clauses.extend(predicate.into_iter()),
}
}
if clauses.is_empty() {
Err(CompilationError::InconsistentEntry)
} else {
Ok(TopLevel::Predicate(clauses))
}
}
fn mark_cut_variables_as(terms: &mut Vec<Term>, name: Atom) {
for term in terms.iter_mut() {
match term {
&mut Term::Literal(_, Literal::Atom(ref mut var)) if *var == atom!("!") => {
*var = name;
}
_ => {}
}
}
}
fn mark_cut_variable(term: &mut Term) -> bool {
let cut_var_found = match term {
&mut Term::Literal(_, Literal::Atom(ref var)) if *var == atom!("!") => true,
_ => false,
};
if cut_var_found {
*term = Term::Var(Cell::default(), Var::from("!"));
true
} else {
false
}
}
fn mark_cut_variables(terms: &mut Vec<Term>) -> bool {
let mut found_cut_var = false;
for item in terms.iter_mut() {
found_cut_var = mark_cut_variable(item) || found_cut_var;
}
found_cut_var
}
// terms is a list of goals composing one clause in a (;) functor. it
// checks that the first (and only) of these clauses is a ->. if so,
// it expands its terms using a blocked_!.
fn check_for_internal_if_then(terms: &mut Vec<Term>) {
if terms.len() != 1 {
return;
}
if let Some(Term::Clause(_, name, ref subterms)) = terms.last() {
if *name != atom!("->") || source_arity(subterms) != 2 {
return;
}
} else {
return;
}
if let Some(Term::Clause(_, _, mut subterms)) = terms.pop() {
let mut conq_terms = VecDeque::from(unfold_by_str(subterms.pop().unwrap(), atom!(",")));
let mut pre_cut_terms = VecDeque::from(unfold_by_str(subterms.pop().unwrap(), atom!(",")));
conq_terms.push_front(Term::Literal(
Cell::default(),
Literal::Atom(atom!("blocked_!")),
));
while let Some(term) = pre_cut_terms.pop_back() {
conq_terms.push_front(term);
}
let tail_term = conq_terms.pop_back().unwrap();
terms.push(fold_by_str(
conq_terms.into_iter(),
tail_term,
atom!(","),
));
}
}
pub(super) fn setup_declaration<'a, LS: LoadState<'a>>(
loader: &mut Loader<'a, LS>,
mut terms: Vec<Term>,
@@ -569,7 +457,7 @@ fn build_meta_predicate_clause<'a, LS: LoadState<'a>>(
}
#[inline]
fn clause_to_query_term<'a, LS: LoadState<'a>>(
pub(super) fn clause_to_query_term<'a, LS: LoadState<'a>>(
loader: &mut Loader<'a, LS>,
name: Atom,
mut terms: Vec<Term>,
@@ -608,7 +496,7 @@ fn clause_to_query_term<'a, LS: LoadState<'a>>(
}
#[inline]
fn qualified_clause_to_query_term<'a, LS: LoadState<'a>>(
pub(super) fn qualified_clause_to_query_term<'a, LS: LoadState<'a>>(
loader: &mut Loader<'a, LS>,
module_name: Atom,
name: Atom,
@@ -646,308 +534,65 @@ fn qualified_clause_to_query_term<'a, LS: LoadState<'a>>(
QueryTerm::Clause(Cell::default(), ct, terms, call_policy)
}
fn compute_head(term: &Term) -> Vec<Term> {
let mut vars = IndexSet::new();
for term in post_order_iter(term) {
if let TermRef::Var(_, _, v) = term {
vars.insert(v.clone());
}
}
vars.insert(Var::from("!"));
vars.into_iter()
.map(|v| Term::Var(Cell::default(), v))
.collect()
}
pub(crate) fn build_rule_body(vars: &[Term], body_term: Term) -> Term {
let head_term = Term::Clause(Cell::default(), atom!(""), vars.iter().cloned().collect());
let rule = vec![head_term, body_term];
Term::Clause(Cell::default(), atom!(":-"), rule)
}
// the terms form the body of the rule. We create a head, by
// gathering variables from the body of terms and recording them
// in the head clause.
fn build_rule(body_term: Term) -> (JumpStub, VecDeque<Term>) {
// collect the vars of body_term into a head, return the num_vars
// (the arity) as well.
let vars = compute_head(&body_term);
let rule = build_rule_body(&vars, body_term);
(vars, VecDeque::from(vec![rule]))
}
fn build_disjunct(body_term: Term) -> (JumpStub, VecDeque<Term>) {
let vars = compute_head(&body_term);
let results = unfold_by_str(body_term, atom!(";"))
.into_iter()
.map(|term| {
let mut subterms = unfold_by_str(term, atom!(","));
mark_cut_variables(&mut subterms);
check_for_internal_if_then(&mut subterms);
let term = subterms.pop().unwrap();
let clause = fold_by_str(subterms.into_iter(), term, atom!(","));
build_rule_body(&vars, clause)
})
.collect();
(vars, results)
}
fn build_if_then(prec: Term, conq: Term) -> (JumpStub, VecDeque<Term>) {
let mut prec_seq = unfold_by_str(prec, atom!(","));
let comma_sym = atom!(",");
let cut_sym = Literal::Atom(atom!("!"));
prec_seq.push(Term::Literal(Cell::default(), cut_sym));
mark_cut_variables_as(&mut prec_seq, atom!("blocked_!"));
let mut conq_seq = unfold_by_str(conq, atom!(","));
mark_cut_variables(&mut conq_seq);
prec_seq.extend(conq_seq.into_iter());
let back_term = prec_seq.pop().unwrap();
let front_term = prec_seq.pop().unwrap();
let body_term = Term::Clause(
Cell::default(),
comma_sym,
vec![front_term, back_term],
);
build_rule(fold_by_str(prec_seq.into_iter(), body_term, comma_sym))
}
#[derive(Debug)]
pub(crate) struct Preprocessor {
queue: VecDeque<VecDeque<Term>>,
settings: CodeGenSettings,
}
impl Preprocessor {
pub(super) fn new(settings: CodeGenSettings) -> Self {
Preprocessor {
queue: VecDeque::new(),
settings,
}
}
fn setup_fact(&mut self, term: Term) -> Result<Term, CompilationError> {
fn setup_fact(&mut self, term: Term) -> Result<Fact, CompilationError> {
match term {
Term::Clause(..) | Term::Literal(_, Literal::Atom(..)) => Ok(term),
Term::Clause(..) | Term::Literal(_, Literal::Atom(..)) => {
let mut classifier = VariableClassifier::new(
self.settings.default_call_policy(),
);
let (head, var_records) = classifier.classify_fact(term)?;
Ok(Fact { head, var_records })
}
_ => Err(CompilationError::InadmissibleFact),
}
}
fn to_query_term<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
term: Term,
) -> Result<QueryTerm, CompilationError> {
match term {
Term::Literal(_, Literal::Atom(name)) => {
if name == atom!("!") || name == atom!("blocked_!") {
Ok(QueryTerm::BlockedCut)
} else {
Ok(clause_to_query_term(
loader,
name,
vec![],
self.settings.default_call_policy(),
))
}
}
Term::Literal(_, Literal::Char('!')) => Ok(QueryTerm::BlockedCut),
Term::Var(_, ref v) if v.as_str() == Some("!") => {
Ok(QueryTerm::UnblockedCut(Cell::default()))
}
Term::Clause(r, name, mut terms) => match (name, source_arity(&terms)) {
(atom!(";"), 2) => {
let term = Term::Clause(r, name, terms);
let (stub, clauses) = build_disjunct(term);
self.queue.push_back(clauses);
Ok(QueryTerm::Jump(stub))
}
(atom!("->"), 2) => {
let conq = terms.pop().unwrap();
let prec = terms.pop().unwrap();
let (stub, clauses) = build_if_then(prec, conq);
self.queue.push_back(clauses);
Ok(QueryTerm::Jump(stub))
}
(atom!("\\+"), 1) => {
terms.push(Term::Literal(
Cell::default(),
Literal::Atom(atom!("$fail")),
));
let conq = Term::Literal(Cell::default(), Literal::Atom(atom!("true")));
let prec = Term::Clause(Cell::default(), atom!("->"), terms);
let terms = vec![prec, conq];
let term = Term::Clause(Cell::default(), atom!(";"), terms);
let (stub, clauses) = build_disjunct(term);
debug_assert!(clauses.len() > 0);
self.queue.push_back(clauses);
Ok(QueryTerm::Jump(stub))
}
(atom!("$get_level"), 1) => {
if let Term::Var(_, ref var) = &terms[0] {
Ok(QueryTerm::GetLevelAndUnify(Cell::default(), var.clone()))
} else {
Err(CompilationError::InadmissibleQueryTerm)
}
}
(atom!(":"), 2) => {
let predicate_name = terms.pop().unwrap();
let module_name = terms.pop().unwrap();
match (module_name, predicate_name) {
(
Term::Literal(_, Literal::Atom(module_name)),
Term::Literal(_, Literal::Atom(predicate_name)),
) => Ok(qualified_clause_to_query_term(
loader,
module_name,
predicate_name,
vec![],
self.settings.default_call_policy(),
)),
(
Term::Literal(_, Literal::Atom(module_name)),
Term::Clause(_, name, terms),
) => Ok(qualified_clause_to_query_term(
loader,
module_name,
name,
terms,
self.settings.default_call_policy()
)),
(module_name, predicate_name) => {
terms.push(module_name);
terms.push(predicate_name);
Ok(clause_to_query_term(
loader,
atom!("call"),
vec![Term::Clause(r, name, terms)],
self.settings.default_call_policy(),
))
}
}
}
_ => Ok(clause_to_query_term(loader, name, terms,
self.settings.default_call_policy())),
},
Term::Var(..) => Ok(QueryTerm::Clause(
Cell::default(),
ClauseType::CallN(1),
vec![term],
self.settings.default_call_policy(),
)),
_ => Err(CompilationError::InadmissibleQueryTerm),
}
}
fn pre_query_term<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
term: Term,
) -> Result<QueryTerm, CompilationError> {
match term {
Term::Clause(r, name, mut subterms) => {
if subterms.len() == 1 && name == atom!("$call_with_inference_counting") {
self.to_query_term(loader, subterms.pop().unwrap())
.map(|mut query_term| {
query_term.set_call_policy(CallPolicy::Counted);
query_term
})
} else {
let clause = Term::Clause(r, name, subterms);
self.to_query_term(loader, clause)
}
}
_ => self.to_query_term(loader, term),
}
}
fn setup_query<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
terms: Vec<Term>,
cut_context: CutContext,
) -> Result<Vec<QueryTerm>, CompilationError> {
let mut query_terms = vec![];
let mut work_queue = VecDeque::from(terms);
while let Some(term) = work_queue.pop_front() {
let mut term = term;
if let Term::Clause(cell, name, terms) = term {
if name == atom!(",") && source_arity(&terms) == 2 {
let term = Term::Clause(cell, name, terms);
let mut subterms = unfold_by_str(term, atom!(","));
while let Some(subterm) = subterms.pop() {
work_queue.push_front(subterm);
}
continue;
} else {
term = Term::Clause(cell, name, terms);
}
}
if let CutContext::HasCutVariable = cut_context {
mark_cut_variable(&mut term);
}
query_terms.push(self.pre_query_term(loader, term)?);
}
Ok(query_terms)
}
fn setup_rule<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
mut terms: Vec<Term>,
cut_context: CutContext,
head: Term,
body: Term,
) -> Result<Rule, CompilationError> {
let post_head_terms: Vec<_> = terms.drain(1..).collect();
let mut query_terms = self.setup_query(loader, post_head_terms, cut_context)?;
let mut classifier = VariableClassifier::new(
self.settings.default_call_policy(),
);
let (head, mut query_terms, var_records) =
classifier.classify_rule(loader, head, body)?;
let clauses = query_terms.drain(1..).collect();
let qt = query_terms.pop().unwrap();
match terms.pop().unwrap() {
match head {
Term::Clause(_, name, terms) => Ok(Rule {
head: (name, terms, qt),
clauses,
var_records,
}),
Term::Literal(_, Literal::Atom(name)) => Ok(Rule {
head: (name, vec![], qt),
clauses,
var_records,
}),
_ => Err(CompilationError::InvalidRuleHead),
}
}
/*
fn try_term_to_query<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
@@ -960,23 +605,19 @@ impl Preprocessor {
cut_context,
)?))
}
*/
pub(super) fn try_term_to_tl<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
term: Term,
cut_context: CutContext,
) -> Result<TopLevel, CompilationError> {
match term {
Term::Clause(r, name, terms) => {
if name == atom!("?-") {
self.try_term_to_query(loader, terms, cut_context)
} else if name == atom!(":-") && terms.len() == 2 {
Ok(TopLevel::Rule(self.setup_rule(
loader,
terms,
cut_context,
)?))
let is_rule = name == atom!(":-") && terms.len() == 2;
if is_rule {
Ok(TopLevel::Rule(self.setup_rule(loader, terms[0], terms[1])?))
} else {
let term = Term::Clause(r, name, terms);
Ok(TopLevel::Fact(self.setup_fact(term)?))
@@ -990,33 +631,13 @@ impl Preprocessor {
&mut self,
loader: &mut Loader<'a, LS>,
terms: I,
cut_context: CutContext,
) -> Result<VecDeque<TopLevel>, CompilationError> {
let mut results = VecDeque::new();
for term in terms.into_iter() {
results.push_back(self.try_term_to_tl(loader, term, cut_context)?);
results.push_back(self.try_term_to_tl(loader, term)?);
}
Ok(results)
}
pub(super) fn parse_queue<'a, LS: LoadState<'a>>(
&mut self,
loader: &mut Loader<'a, LS>,
) -> Result<VecDeque<TopLevel>, CompilationError> {
let mut queue = VecDeque::new();
while let Some(terms) = self.queue.pop_front() {
let clauses = merge_clauses(&mut self.try_terms_to_tls(
loader,
terms,
CutContext::HasCutVariable,
)?)?;
queue.push_back(clauses);
}
Ok(queue)
}
}