//! This module implements a zero-copy version of the runtime parser that
//! uses the LR statemachine generated by rustlr. It will (for now), live
//! side by side with the original parser implemented as [crate::RuntimeParser].
//!
//! This module implements the parsing routines that uses the state machine
//! generated by rustlr. **The main structure here is [ZCParser]**.
//! All parsing functions are organized around the [ZCParser::parse_core]
//! function, which implements the basic LR parsing algorithm. This function
//! expects dynamic [Tokenizer] and [ErrReporter] trait-objects.
//! This module provides generic
//! parsing and parser-training routines that use stdio for interface, but
//! the [ErrReporter] trait allows custom user interfaces to be build separately.
#![allow(dead_code)]
#![allow(unused_variables)]
#![allow(non_snake_case)]
#![allow(non_camel_case_types)]
#![allow(unused_parens)]
#![allow(unused_mut)]
#![allow(unused_assignments)]
#![allow(unused_doc_comments)]
#![allow(unused_imports)]
use std::fmt::Display;
use std::default::Default;
use std::collections::{HashMap,HashSet,BTreeSet};
use std::io::{self,Read,Write,BufReader,BufRead};
use std::cell::{RefCell,Ref,RefMut};
use std::hash::{Hash,Hasher};
use std::any::Any;
use std::fs::File;
use std::io::prelude::*;
use std::path::Path;
use std::mem;
use crate::{TRACE,Stateaction,Statemachine,TerminalToken,Tokenizer};
use crate::{LBox,LRc};
use crate::Stateaction::*;
use std::rc::Rc;
use crate::{StandardReporter};
/// this structure is only exported because it is required by the generated parsers.
/// There is no reason to use it in other programs. Replaces [crate::RProduction] for new parsers since version 0.2.0
#[derive(Clone)]
pub struct ZCRProduction<AT:Default,ET:Default> // runtime rep of grammar rule
{
pub lhs: &'static str, // left-hand side nonterminal of rule
pub Ruleaction : fn(&mut ZCParser<AT,ET>) -> AT, //parser as arg
}
impl<AT:Default,ET:Default> ZCRProduction<AT,ET>
{
pub fn new_skeleton(lh:&'static str) -> ZCRProduction<AT,ET>
{
ZCRProduction {
lhs : lh,
Ruleaction : |p|{/*println!("EXECUTING DEFAULT!!");*/ <AT>::default()},
}
}
}//impl ZCRProduction
/// these structures are what's on the parse stack. When writing a
/// grammar rule such as `E --> E:a + E:b`, the variables a and b will
/// be bound to values of this type, and thus inside the semantic actions
/// one would need to use `a.value` to extract the value, which is of the
/// declared 'absyntype' of the grammar. Alternatively, one can use a
/// pattern: `E:@a@ + E:@b@` to bind a and b directly to the values.
pub struct StackedItem<AT:Default> // replaces Stackelement
{
si : usize, // state index
pub value : AT, // semantic value (don't clone grammar symbols)
pub line: usize, // line and column
pub column: usize,
}
impl<AT:Default> StackedItem<AT>
{
pub fn new(si:usize,value:AT,line:usize,column:usize) -> StackedItem<AT>
{ StackedItem{si,value,line,column} }
/// converts the information in a stacked item to an [LBox] enclosing
/// the line and column numbers
pub fn lbox(self) -> LBox<AT>
{ LBox::new(self.value,self.line,self.column) }
}
/// this is the structure created by the generated parser. The generated parser
/// program will contain a make_parser function that returns this structure.
/// Most of the pub items are, however, only exported to support the operation
/// of the parser, and should not be accessed directly. Only the functions
/// [ZCParser::parse], [ZCParser::report], [ZCParser::abort]
/// and [ZCParser::error_occurred] should be called directly
/// from user programs. Only the field [ZCParser::exstate] should be accessed
/// by user programs.
pub struct ZCParser<AT:Default,ET:Default>
{
/// this is the "external state" structure, with type ET defined by the grammar.
/// The semantic actions associated with each grammar rule, which are written
/// in the grammar, have ref mut access to the ZCParser structure, which
/// allows them to read and change the external state object. This gives
/// the parsers greater flexibility and capability, including the ability to
/// parse some non-context free languages. See
/// [this sample grammar](<https://cs.hofstra.edu/~cscccl/rustlr_project/ncf.grammar>).
/// The exstate is initialized to ET::default().
pub exstate : ET, // external state structure, usage optional
/// used only by generated parser: do not reference
pub RSM : Vec<HashMap<&'static str,Stateaction>>, // runtime state machine
// do not reference
//pub Expected : Vec<Vec<&'static str>>,
/// do not reference
pub Rules : Vec<ZCRProduction<AT,ET>>, //rules with just lhs and delegate function
////// this value should be set through abort or report
stopparsing : bool,
/// do not reference
pub stack : Vec<StackedItem<AT>>, // parse stack
// pub recover : HashSet<&'static str>, // for error recovery
pub resynch : HashSet<&'static str>,
pub Errsym : &'static str,
err_occurred : bool,
/// axiom: linenum and column represents the starting position of the
/// topmost StackedItem.
pub linenum : usize,
pub column : usize,
// pub src_id : usize,
report_line : usize,
/// Hashset containing all grammar symbols (terminal and non-terminal). This is used for error reporting and training.
pub Symset : HashSet<&'static str>,
//pub tokenizer:&'t mut dyn Tokenizer<'t,AT>,
popped : Vec<(usize,usize)>,
}//struct ZCParser
impl<AT:Default,ET:Default> ZCParser<AT,ET>
{
/// this is only called by the make_parser function in the machine-generated
/// parser program. *Do not call this function in other places* as it
/// only generates a skeleton.
pub fn new(rlen:usize, slen:usize/*,tk:&'t mut dyn Tokenizer<'t,AT>*/) -> ZCParser<AT,ET>
{ // given number of rules and number states
let mut p = ZCParser {
RSM : Vec::with_capacity(slen),
//Expected : Vec::with_capacity(slen),
Rules : Vec::with_capacity(rlen),
stopparsing : false,
exstate : ET::default(),
stack : Vec::with_capacity(1024),
Errsym : "",
err_occurred : false,
linenum : 0,
column : 0,
// src_id : 0,
report_line : 0,
resynch : HashSet::new(),
//added for training
//training : false,
//trained : HashMap::new(),
Symset : HashSet::with_capacity(64),
//tokenizer:tk,
popped: Vec::with_capacity(8),
};
for _ in 0..slen {
p.RSM.push(HashMap::with_capacity(16));
//p.Expected.push(Vec::new());
}
return p;
}//new
/// this function can be called from with the "semantic" actions attached
/// to grammar production rules that are executed for each
/// "reduce" action of the parser.
pub fn abort(&mut self, msg:&str)
{
eprintln!("\n!!!Parsing Aborted: {}",msg);
self.err_occurred = true;
self.stopparsing=true;
}
/// may be called from grammar semantic actions to report error.
/// this report function will print to stdout.
pub fn report(&mut self, errmsg:&str)
{ // linenum must be set prior to call
if (self.report_line != self.linenum || self.linenum==0) {
// eprint!("ERROR on line {}, column {}:\n{}\n",self.linenum,self.column,tokenizer.current_line());
// eprintln!("ERROR on line {}, column {}: {}",self.linenum,self.column,errmsg);
eprintln!("PARSER ERROR: {}",errmsg);
self.report_line = self.linenum;
}
else {
eprint!(" {} ",errmsg);
}
self.err_occurred = true;
}// report
/// this function is only exported to support the generated code
pub fn bad_pattern(&mut self,pattern:&str) -> AT
{
let msg = format!("pattern {} failed to bind to stacked values\n",pattern);
self.report(&msg);
//println!("FROM BAD PATTERN:");
AT::default()
}
/*
/// sets an index that index source information, such as the source file
/// when compiling multiple sources. This information must be maintained externally.
/// The source id will also be passed on to the [LBox] and [LRc] smartpointers by
/// the [ZCParser::lb] function.
pub fn set_src_id(&mut self, id:usize)
{ self.src_id =id; }
*/
//called to simulate a shift
fn errshift(&mut self, sym:&str) -> bool
{
let csi = self.stack[self.stack.len()-1].si; // current state
let actionopt = self.RSM[csi].get(sym);
if let Some(Shift(ni)) = actionopt {
self.stack.push(StackedItem::new(*ni,AT::default(),self.linenum,self.column)); true
//self.stack.push(Stackelement{si:*ni,value:AT::default()}); true
}
else {false}
}
// this is the LR parser shift action: push the next state, along with the
// value of the current lookahead token onto the parse stack, returns the
// next token
fn shift<'t>(&mut self, nextstate:usize, lookahead:TerminalToken<'t,AT>, tokenizer:&mut dyn Tokenizer<'t,AT>) -> TerminalToken<'t, AT>
{
//self.stack.push(Stackelement{si:nextstate,value:lookahead.value});
self.linenum = lookahead.line; self.column=lookahead.column;
self.stack.push(StackedItem::new(nextstate,lookahead.value,lookahead.line,lookahead.column));
//self.nexttoken()
tokenizer.next_tt()
}
/// this function is called from the generated semantic actions and should
/// most definitely not be called from elsewhere as it would corrupt
/// the base parser.
pub fn popstack(&mut self) -> StackedItem<AT>
{
let item = self.stack.pop().expect("PARSER STATE MACHINE/STACK CORRUPTED");
self.linenum = item.line; self.column=item.column;
self.popped.push((item.line,item.column));
item
}//popstack
fn reduce(&mut self, ri:&usize)
{
self.popped.clear();
let rulei = &self.Rules[*ri];
let ruleilhs = rulei.lhs; // &'static : Copy
//let mut dummy = RuntimeParser::new(1,1);
let val = (rulei.Ruleaction)(self); // should be self
let newtop = self.stack[self.stack.len()-1].si;
let goton = self.RSM[newtop].get(ruleilhs).expect("PARSER STATEMACHINE CORRUPTED");
if let Stateaction::Gotonext(nsi) = goton {
/*
the line/column must be the last thing that was popped, but how is this communicated from the semantic actions?
Solution: When the semantic action pops, it changes self.linenum,self.column,
instead of pop, there should be a function self.popstack() that returns value.
This is correct because linenum/column will again reflect start of tos item
*/
self.stack.push(StackedItem::new(*nsi,val,self.linenum,self.column));
//self.stack.push(Stackelement{si:*nsi,value:val});
}// goto next state after reduce
else {
self.report("state transition table corrupted: no suitable action after reduce");
self.stopparsing=true;
}
}//reduce
/// can be called to determine if an error occurred during parsing. The parser
/// will not panic.
pub fn error_occurred(&self) -> bool {self.err_occurred}
// there may need to be other lb functions, perhaps from terminalToken
// or stackedItem (at least for transfer)
/// creates a [LBox] smart pointer that includes line/column information;
/// should be called from the semantic actions of a grammar rule, e.g.
///```ignore
/// E --> E:a + E:b {PlusExpr(parser.lb(a),parser.lb(b))}
///```
pub fn lb<T>(&self,e:T) -> LBox<T> { LBox::new(e,self.linenum,self.column /*,self.src_id*/) }
/// creates a `LBox<dyn Any>`, which allows attributes of different types to
/// be associated with grammar symbols. Use in conjuction with [LBox::downcast], [LBox::upcast] and the [lbdown], [lbup] macros.
pub fn lba<T:'static>(&self,e:T) -> LBox<dyn Any> { LBox::upcast(LBox::new(e,self.linenum,self.column /*,self.src_id*/)) }
/// similar to [ZCParser::lb], but creates a [LRc] instead of [LBox]
pub fn lrc<T>(&self,e:T) -> LRc<T> { LRc::new(e,self.linenum,self.column /*,self.src_id*/) }
/// similar to [ZCParser::lba] but creates a [LRc]
pub fn lrca<T:'static>(&self,e:T) -> LRc<dyn Any> { LRc::upcast(LRc::new(e,self.linenum,self.column /*,self.src_id*/)) }
/// creates LBox enclosing e using line/column information associated
/// with right-hand side symbols, numbered left-to-right starting at 0
pub fn lbx<T>(&self,i:usize,e:T) -> LBox<T>
{
let (mut ln,mut cl) = (self.linenum,self.column);
if i<self.popped.len() {
let index = self.popped.len() - 1 - i;
let lc = self.popped[index];
ln = lc.0; cl=lc.1;
}
LBox::new(e,ln,cl)
}//lbx
/// Like lbx but creates an LRc
pub fn lrcn<T>(&self,i:usize,e:T) -> LRc<T>
{
let (mut ln,mut cl) = (self.linenum,self.column);
if i<self.popped.len() {
let index = self.popped.len() - 1 - i;
let lc = self.popped[index];
ln = lc.0; cl=lc.1;
}
LRc::new(e,ln,cl)
}//lbx
}// impl ZCParser
//////////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////////
//// new version of write_fsm:
impl Statemachine
{ /////// zc version
pub fn writezcparser(&self, filename:&str)->Result<(),std::io::Error>
{
let ref absyn = self.Gmr.Absyntype;
let ref extype = self.Gmr.Externtype;
let ref lifetime = self.Gmr.lifetime;
let has_lt = lifetime.len()>0 && (absyn.contains(lifetime) || extype.contains(lifetime));
let ltopt = if has_lt {format!("<{}>",lifetime)} else {String::from("")};
let mut fd = File::create(filename)?;
write!(fd,"//Parser generated by rustlr for grammar {}",&self.Gmr.name)?;
write!(fd,"\n
#![allow(unused_variables)]
#![allow(non_snake_case)]
#![allow(non_camel_case_types)]
#![allow(unused_parens)]
#![allow(unused_mut)]
#![allow(unused_imports)]
#![allow(unused_assignments)]
#![allow(dead_code)]
#![allow(irrefutable_let_patterns)]
extern crate rustlr;
use rustlr::{{Tokenizer,TerminalToken,ZCParser,ZCRProduction,Stateaction,decode_action}};\n")?;
write!(fd,"{}\n",&self.Gmr.Extras)?; // use clauses and such
// write static array of symbols
write!(fd,"const SYMBOLS:[&'static str;{}] = [",self.Gmr.Symbols.len())?;
for i in 0..self.Gmr.Symbols.len()-1
{
write!(fd,"\"{}\",",&self.Gmr.Symbols[i].sym)?;
}
write!(fd,"\"{}\"];\n\n",&self.Gmr.Symbols[self.Gmr.Symbols.len()-1].sym)?;
// position of symbols must be inline with self.Gmr.Symhash
// record table entries in a static array
let mut totalsize = 0;
for i in 0..self.FSM.len() { totalsize+=self.FSM[i].len(); }
write!(fd,"const TABLE:[u64;{}] = [",totalsize)?;
// generate table to represent FSM
let mut encode:u64 = 0;
for i in 0..self.FSM.len() // for each state index i
{
let row = &self.FSM[i];
for key in row.keys()
{ // see function decode for opposite translation
let k = *self.Gmr.Symhash.get(key).unwrap(); // index of symbol
encode = ((i as u64) << 48) + ((k as u64) << 32);
match row.get(key) {
Some(Shift(statei)) => { encode += (*statei as u64) << 16; },
Some(Gotonext(statei)) => { encode += ((*statei as u64) << 16)+1; },
Some(Reduce(rulei)) => { encode += ((*rulei as u64) << 16)+2; },
Some(Accept) => {encode += 3; },
_ => {encode += 4; }, // 4 indicates Error
}//match
write!(fd,"{},",encode)?;
} //for symbol index k
}//for each state index i
write!(fd,"];\n\n")?;
// must know what absyn type is when generating code.
write!(fd,"pub fn new_parser{}() -> ZCParser<{},{}>",<opt,absyn,extype)?;
write!(fd,"\n{{\n")?;
// write code to pop stack, assign labels to variables.
write!(fd," let mut parser1:ZCParser<{},{}> = ZCParser::new({},{});\n",absyn,extype,self.Gmr.Rules.len(),self.States.len())?;
// generate rules and Ruleaction delegates, must pop values from runtime stack
write!(fd," let mut rule = ZCRProduction::<{},{}>::new_skeleton(\"{}\");\n",absyn,extype,"start")?;
for i in 0..self.Gmr.Rules.len()
{
write!(fd," rule = ZCRProduction::<{},{}>::new_skeleton(\"{}\");\n",absyn,extype,self.Gmr.Rules[i].lhs.sym)?;
write!(fd," rule.Ruleaction = |parser|{{ ")?;
let mut k = self.Gmr.Rules[i].rhs.len();
//form if-let labels and patterns as we go...
let mut labels = String::from("(");
let mut patterns = String::from("(");
while k>0 // k is length of right-hand side
{
let gsym = &self.Gmr.Rules[i].rhs[k-1];
let findat = gsym.label.find('@');
let mut plab = format!("_item{}_",k-1);
match &findat {
None if gsym.label.len()>0 && !gsym.label.contains('(') => {
plab=format!("{}",gsym.label.trim());
},
Some(ati) if *ati>0 => {plab=format!("{}",&gsym.label[0..*ati]);},
_ => {},
}//match
let poppedlab = plab.as_str();
write!(fd,"let mut {} = parser.popstack(); ",poppedlab)?;
if gsym.label.len()>1 && findat.is_some() { // if-let pattern
let atindex = findat.unwrap();
if atindex>0 { // label like es:@Exp(..)@
//let varlab = &gsym.label[0..atindex]; //es before @: es:@..@
labels.push_str("&mut "); // for if-let
labels.push_str(poppedlab); labels.push_str(".value,");
//write!(fd," let mut {}={}.value; ",varlab,poppedlab)?;
}
else { // non-labeled pattern: E:@..@
labels.push_str(poppedlab); labels.push_str(".value,");
}
patterns.push_str(&gsym.label[atindex+1..]); patterns.push(',');
} // @@ pattern exists, with or without label
else if gsym.label.len()>0 && gsym.label.contains('(') // simple label like E:(a,b)
{ // label exists but only simple pattern
labels.push_str(poppedlab); labels.push_str(".value,");
patterns.push_str(&gsym.label[..]); // non-mutable
patterns.push(',')
}// simple label
// else simple label is not a pattern, so do nothing
k -= 1;
}// for each symbol on right hand side of rule
// form if let pattern=labels ...
let defaultaction = format!("<{}>::default()}}",absyn);
let mut semaction = &self.Gmr.Rules[i].action; //string that ends with }
if semaction.len()<=1 {semaction = &defaultaction;}
if labels.len()<2 { write!(fd,"{};\n",semaction.trim_end())?; } //empty pattern
else { // write an if-let
labels.push(')'); patterns.push(')');
write!(fd,"\n if let {}={} {{ {} else {{parser.bad_pattern(\"{}\")}} }};\n",&patterns,&labels,semaction.trim_end(),&patterns)?;
}// if-let semantic action
write!(fd," parser1.Rules.push(rule);\n")?;
}// for each rule
write!(fd," parser1.Errsym = \"{}\";\n",&self.Gmr.Errsym)?;
// resynch vector
for s in &self.Gmr.Resynch {write!(fd," parser1.resynch.insert(\"{}\");\n",s)?;}
// generate code to load RSM from TABLE
write!(fd,"\n for i in 0..{} {{\n",totalsize)?;
write!(fd," let symi = ((TABLE[i] & 0x0000ffff00000000) >> 32) as usize;\n")?;
write!(fd," let sti = ((TABLE[i] & 0xffff000000000000) >> 48) as usize;\n")?;
write!(fd," parser1.RSM[sti].insert(SYMBOLS[symi],decode_action(TABLE[i]));\n }}\n\n")?;
// write!(fd,"\n for i in 0..{} {{for k in 0..{} {{\n",rows,cols)?;
// write!(fd," parser1.RSM[i].insert(SYMBOLS[k],decode_action(TABLE[i*{}+k]));\n }}}}\n",cols)?;
write!(fd," for s in SYMBOLS {{ parser1.Symset.insert(s); }}\n\n")?;
write!(fd," load_extras(&mut parser1);\n")?;
write!(fd," return parser1;\n")?;
write!(fd,"}} //make_parser\n\n")?;
////// Augment!
write!(fd,"fn load_extras{}(parser:&mut ZCParser<{},{}>)\n{{\n",<opt,absyn,extype)?;
write!(fd,"}}//end of load_extras: don't change this line as it affects augmentation\n")?;
Ok(())
}//writezcparser
/////////////////////LBA VERSION//////////////////////////////////////
///// semantic acition fn is _semaction_for_{rule index}
////////////////////////////////////////////////
//////////////////////////// write parser for LBox<dyn Any>
pub fn writelbaparser(&self, filename:&str)->Result<(),std::io::Error>
{
let ref absyn = self.Gmr.Absyntype;
if !is_lba(absyn) /*absyn!="LBox<dyn Any>" && absyn!="LBox<Any>"*/ {
return self.writezcparser(filename);
}
let ref extype = self.Gmr.Externtype;
let ref lifetime = self.Gmr.lifetime;
let has_lt = lifetime.len()>0 && (absyn.contains(lifetime) || extype.contains(lifetime));
let ltopt = if has_lt {format!("<{}>",lifetime)} else {String::from("")};
let rlen = self.Gmr.Rules.len();
// generate action fn's from strings stored in gen-time grammar
let mut actions:Vec<String> = Vec::with_capacity(rlen);
for ri in 0..rlen
{
let lhs = &self.Gmr.Rules[ri].lhs.sym;
let lhsi = self.Gmr.Symhash.get(lhs).expect("GRAMMAR REPRESENTATION CORRUPTED");
let rettype = &self.Gmr.Symbols[*lhsi].rusttype; // return type
let ltoptr = if has_lt || (lifetime.len()>0 && rettype.contains(lifetime))
{format!("<{}>",lifetime)} else {String::from("")};
let mut fndef = format!("fn _semaction_for_{}_{}(parser:&mut ZCParser<{},{}>) -> {} {{\n",ri,<optr,absyn,extype,rettype);
let mut k = self.Gmr.Rules[ri].rhs.len();
//form if-let labels and patterns as we go...
let mut labels = String::from("(");
let mut patterns = String::from("(");
while k>0 // k is length of right-hand side
{
let gsym = &self.Gmr.Rules[ri].rhs[k-1]; // rhs symbols right to left...
let gsymi = *self.Gmr.Symhash.get(&gsym.sym).unwrap();
let findat = gsym.label.find('@');
let mut plab = format!("_item{}_",k-1);
match &findat {
None if gsym.label.len()>0 => {plab = format!("{}",&gsym.label);},
Some(ati) if *ati>0 => {plab=format!("{}",&gsym.label[0..*ati]);},
_ => {},
}//match
let poppedlab = plab.as_str();
let ref symtype = gsym.rusttype;
let mut stat = format!("let mut {} = lbdown!(parser.popstack().value,{}); ",poppedlab,symtype); // no longer stackitem but lbdown!
if symtype.len()<2 || symtype=="LBox<dyn Any>" || symtype=="LBox<Any>" {
stat = format!("let mut {} = parser.popstack().value; ",poppedlab);
// no need for lbdown if type is already LBA
}
fndef.push_str(&stat);
// poppedlab now bound to lbdown!
if gsym.label.len()>1 && findat.is_some() { // if-let pattern
labels.push_str("&mut *"); // for if-let // *box.exp gets value
labels.push_str(poppedlab); /*labels.push_str(".exp");*/ labels.push(',');
// closing @ trimed in grammar_processor.rs
let atindex = findat.unwrap();
patterns.push_str(&gsym.label[atindex+1..]); patterns.push(',');
} // @@ pattern exists, with or without label
k -= 1;
}// for each symbol on right hand side of rule (while k)
// form if let pattern=labels ...
let defaultaction = format!("<{}>::default()}}",rettype);
let mut semaction = &self.Gmr.Rules[ri].action; //string that ends w/ rbr
if semaction.len()<=1 {semaction = &defaultaction;}
if labels.len()<2 {
fndef.push_str(semaction.trim_end()); fndef.push_str("\n");
} //empty pattern
else { // write an if-let
labels.push(')'); patterns.push(')');
let pat2= format!("\n if let {}={} {{ {} else {{parser.report(\"{}\"); <{}>::default()}} }}\n",&patterns,&labels,semaction.trim_end(),&patterns,rettype);
fndef.push_str(&pat2);
}// if-let semantic action
actions.push(fndef);
}// generate action function for each rule (for ri..
////// write to file
let mut fd = File::create(filename)?;
write!(fd,"//Parser generated by rustlr for grammar {}",&self.Gmr.name)?;
write!(fd,"\n
#![allow(unused_variables)]
#![allow(non_snake_case)]
#![allow(non_camel_case_types)]
#![allow(unused_parens)]
#![allow(unused_mut)]
#![allow(unused_imports)]
#![allow(unused_assignments)]
#![allow(dead_code)]
#![allow(irrefutable_let_patterns)]
use std::any::Any;
extern crate rustlr;
use rustlr::{{Tokenizer,TerminalToken,ZCParser,ZCRProduction,Stateaction,decode_action,LBox,lbdown,lbup,lbget,unbox}};\n")?;
write!(fd,"{}\n",&self.Gmr.Extras)?; // use clauses and such
// write static array of symbols
write!(fd,"const SYMBOLS:[&'static str;{}] = [",self.Gmr.Symbols.len())?;
for i in 0..self.Gmr.Symbols.len()-1
{
write!(fd,"\"{}\",",&self.Gmr.Symbols[i].sym)?;
}
write!(fd,"\"{}\"];\n\n",&self.Gmr.Symbols[self.Gmr.Symbols.len()-1].sym)?;
// position of symbols must be inline with self.Gmr.Symhash
// record table entries in a static array
let mut totalsize = 0;
for i in 0..self.FSM.len() { totalsize+=self.FSM[i].len(); }
write!(fd,"const TABLE:[u64;{}] = [",totalsize)?;
// generate table to represent FSM
let mut encode:u64 = 0;
for i in 0..self.FSM.len() // for each state index i
{
let row = &self.FSM[i]; ////////LBA VERSION
for key in row.keys()
{ // see function decode for opposite translation
let k = *self.Gmr.Symhash.get(key).unwrap(); // index of symbol
encode = ((i as u64) << 48) + ((k as u64) << 32);
match row.get(key) {
Some(Shift(statei)) => { encode += (*statei as u64) << 16; },
Some(Gotonext(statei)) => { encode += ((*statei as u64) << 16)+1; },
Some(Reduce(rulei)) => { encode += ((*rulei as u64) << 16)+2; },
Some(Accept) => {encode += 3; },
_ => {encode += 4; }, // 4 indicates Error
}//match
write!(fd,"{},",encode)?;
} //for symbol index k
}//for each state index i
write!(fd,"];\n\n")?;
// write action functions
for deffn in &actions { write!(fd,"{}",deffn)?; }
// must know what absyn type is when generating code.
write!(fd,"\npub fn create_parser{}() -> ZCParser<{},{}>",<opt,absyn,extype)?;
write!(fd,"\n{{\n")?;
// write code to pop stack, assign labels to variables.
write!(fd," let mut parser1:ZCParser<{},{}> = ZCParser::new({},{});\n",absyn,extype,self.Gmr.Rules.len(),self.States.len())?;
// generate rules and Ruleaction delegates to call action fns
write!(fd," let mut rule = ZCRProduction::<{},{}>::new_skeleton(\"{}\");\n",absyn,extype,"start")?; // dummy for init
for i in 0..self.Gmr.Rules.len()
{
write!(fd," rule = ZCRProduction::<{},{}>::new_skeleton(\"{}\");\n",absyn,extype,self.Gmr.Rules[i].lhs.sym)?;
write!(fd," rule.Ruleaction = |parser|{{ ")?;
// write code to call action function, then enclose in lba
let lhsi = self.Gmr.Symhash.get(&self.Gmr.Rules[i].lhs.sym).expect("GRAMMAR REPRESENTATION CORRUPTED");
let fnname = format!("_semaction_for_{}_",i);
let typei = &self.Gmr.Symbols[*lhsi].rusttype;
if is_lba(typei) {
write!(fd," {}(parser) }};\n",&fnname)?;
}
else {
write!(fd," lbup!( LBox::new({}(parser),parser.linenum,parser.column)) }};\n",&fnname)?;
}
write!(fd," parser1.Rules.push(rule);\n")?;
}// write each rule action
write!(fd," parser1.Errsym = \"{}\";\n",&self.Gmr.Errsym)?;
// resynch vector
for s in &self.Gmr.Resynch {write!(fd," parser1.resynch.insert(\"{}\");\n",s)?;}
// generate code to load RSM from TABLE
write!(fd,"\n for i in 0..{} {{\n",totalsize)?;
write!(fd," let symi = ((TABLE[i] & 0x0000ffff00000000) >> 32) as usize;\n")?;
write!(fd," let sti = ((TABLE[i] & 0xffff000000000000) >> 48) as usize;\n")?;
write!(fd," parser1.RSM[sti].insert(SYMBOLS[symi],decode_action(TABLE[i]));\n }}\n\n")?;
// write!(fd,"\n for i in 0..{} {{for k in 0..{} {{\n",rows,cols)?;
// write!(fd," parser1.RSM[i].insert(SYMBOLS[k],decode_action(TABLE[i*{}+k]));\n }}}}\n",cols)?;
write!(fd," for s in SYMBOLS {{ parser1.Symset.insert(s); }}\n\n")?;
write!(fd," load_extras(&mut parser1);\n")?;
write!(fd," return parser1;\n")?;
write!(fd,"}} //make_parser\n\n")?;
////// Augment!
write!(fd,"fn load_extras{}(parser:&mut ZCParser<{},{}>)\n{{\n",<opt,absyn,extype)?;
write!(fd,"}}//end of load_extras: don't change this line as it affects augmentation\n")?;
Ok(())
}//writelbaparser
//write-verbose no longer supported
} // impl Statemachine
////// independent function
fn iserror(actionopt:&Option<&Stateaction>) -> bool
{
match actionopt {
None => true,
Some(Error(_)) => true,
_ => false,
}
}//iserror
////// independent function
fn is_lba(t:&str) -> bool {
t.contains("LBox") && t.contains("Any") && t.contains('<') && t.contains('>')
// for s in ["", "LBox<dyn Any>","LBox<Any>","LBox< dyn Any>","LBox<dyn Any >",
// "LBox< dyn Any >"] { if s==t {return true;}}
// return false;
}//is_lba to check type
///////////////////////////////////////////////////////////////////////////
////// reimplementing the parsing algorithm more modularly, with aim of
////// allowing custom parsers
//////////// errors should compile a report
impl<AT:Default,ET:Default> ZCParser<AT,ET>
{
/// Error recovery routine of rustlr, separate from error_reporter.
/// This function will modify the parser and lookahead symbol and return
/// either the next action the parser should take (if recovery succeeded)
/// or None if recovery failed.
pub fn error_recover<'t>(&mut self, lookahead:&mut TerminalToken<'t,AT>, tokenizer:&mut dyn Tokenizer<'t,AT>) -> Option<Stateaction>
{
let mut erraction = None;
///// prefer to ue Errsym method
if self.Errsym.len()>0 {
let errsym = self.Errsym;
// lookdown stack for state with trainsiton on Errsym
// but that could be current state too (start at top)
let mut k = self.stack.len(); // offset by 1 because of usize
let mut spos = k+1;
while k>0 && spos>k
{
let ksi = self.stack[k-1].si;
erraction = self.RSM[ksi].get(errsym);
if let None = erraction {k-=1;} else {spos=k;}
}//while k>0
if spos==k { self.stack.truncate(k); } // new current state revealed
// run all reduce actions that are valid before the Errsym:
while let Some(Reduce(ri)) = erraction // keep reducing
{
//self.reduce(ri); // borrow error- only need mut self.stack
self.popped.clear();
let rulei = &self.Rules[*ri];
let ruleilhs = rulei.lhs; // &'static : Copy
//let mut dummy = RuntimeParser::new(1,1);
let val = (rulei.Ruleaction)(self);
let newtop = self.stack[self.stack.len()-1].si;
let gotonopt = self.RSM[newtop].get(ruleilhs);
match gotonopt {
Some(Gotonext(nsi)) => {
//self.stack.push(Stackelement{si:*nsi,value:val});
self.stack.push(StackedItem::new(*nsi,val,self.linenum,self.column));
},// goto next state after reduce
_ => {self.abort("recovery failed"); },
}//match
// end reduce
let tos=self.stack[self.stack.len()-1].si;
erraction = self.RSM[tos].get(self.Errsym).clone();
} // while let erraction is reduce
// remaining defined action on Errsym must be shift
if let Some(Shift(i)) = erraction { // simulate shift errsym
//self.stack.push(Stackelement{si:*i,value:AT::default()});
self.stack.push(StackedItem::new(*i,AT::default(),lookahead.line,lookahead.column));
// keep lookahead until action is found that transitions from
// current state (i). but skipping ahead without reducing
// the error production is not a good idea
while let None = self.RSM[*i].get(lookahead.sym) {
if lookahead.sym=="EOF" {break;}
*lookahead = tokenizer.next_tt();
}//while let
// either at end of input or found action on next symbol
erraction = self.RSM[*i].get(lookahead.sym);
} // if shift action found down under stack
}//errsym exists
// at this point, if erraction is None, then Errsym failed to recover,
// try the resynch symbol method next ...
if iserror(&erraction) && self.resynch.len()>0 {
while lookahead.sym!="EOF" &&
!self.resynch.contains(lookahead.sym) {
self.linenum = lookahead.line; self.column = lookahead.column;
*lookahead = tokenizer.next_tt();
}//while
if lookahead.sym!="EOF" {
// look for state on stack that has action defined on next symbol
self.linenum = lookahead.line; self.column = lookahead.column;
*lookahead = tokenizer.next_tt();
}
let mut k = self.stack.len()-1; // offset by 1 because of usize
let mut position = 0;
while k>0 && erraction==None
{
let ksi = self.stack[k-1].si;
erraction = self.RSM[ksi].get(lookahead.sym);
if let None=erraction {k-=1;}
}//while k>0 && erraction==None
match erraction {
None => {}, // do nothing, whill shift next symbol
_ => { self.stack.truncate(k);},//pop stack
}//match
}// there are resync symbols
// at this point, if erraction is None, then resynch recovery failed too.
// only action left is to skip ahead...
let mut eofcx = 0;
while iserror(&erraction) && eofcx<1 { //skip input
self.linenum = lookahead.line; self.column = lookahead.column;
*lookahead = tokenizer.next_tt();
//*lookahead = self.nexttoken();
if lookahead.sym=="EOF" {eofcx+=1;}
let csi =self.stack[self.stack.len()-1].si;
erraction = self.RSM[csi].get(lookahead.sym);
}// skip ahead
match erraction {
None => None,
Some(act) => Some(*act),
}//return match
}//error_recover function
}//impl ZCParser 2
/////////////////////////////////////////////////////////////////////////
/////////////// new approach using more flexible trait object
/// A trait object that implements ErrReporter is expected by the [ZCParser::parse_core]
/// function, which implements the basic LR parsing algorithm using the
/// generated state machine. The struct [StandardReporter] is provided as
/// the default ErrReporter that uses standard I/O as interface and has the
/// ability to train the parser. But other implementations of the trait
/// can be created that use different interfaces, such as a graphical IDE.
///
/// This trait replaces [crate::ErrHandler] in the [crate::runtime_parser] module.
pub trait ErrReporter<AT:Default,ET:Default> // not same as error recovery
{
fn err_reporter(&mut self, parser:&mut ZCParser<AT,ET>, lookahead:&TerminalToken<AT>, erropt:&Option<Stateaction>);
fn report_err(&self, parser:&mut ZCParser<AT,ET>, msg:&str) { parser.report(msg) }
// fn training_mode(&self, parser:&ZCParser<AT,ET>) -> bool {false}
// fn interactive_mode(&self, parser:&ZCParser<AT,ET>) -> bool {false}
}// ErrReporter trait // not same as RuntimeParser::ErrHandler
/*
The structure here is a bit strange. The script file is written to in
interactive training mode and read from in script-training mode. However,
the actual modification of the parser file is done after the training, by
the augmenter module. Thus there's another wrapper function that's needed
besides the creation of the right kind of StandardReporter.
*/
impl<AT:Default,ET:Default> ErrReporter<AT,ET> for StandardReporter
{
// this function will be able to write training script to file
fn err_reporter(&mut self, parser:&mut ZCParser<AT,ET>, lookahead:&TerminalToken<AT>, erropt:&Option<Stateaction>)
{
let mut wresult:std::io::Result<()> = Err(std::io::Error::new(std::io::ErrorKind::Other,"")); // dummy
// known that actionop is None or Some(Error(_))
let cstate = parser.stack[parser.stack.len()-1].si; // current state
let mut actionopt = if let Some(act)=erropt {Some(act)} else {None};
let lksym = &lookahead.sym[..];
// is lookahead recognized as a grammar symbol?
// if actionopt is NONE, check entry for ANY_ERROR
if parser.Symset.contains(lksym) {
if let None=actionopt {
actionopt = parser.RSM[cstate].get("ANY_ERROR");
}
}// lookahead is recognized grammar sym
else {
actionopt = parser.RSM[cstate].get("ANY_ERROR");
}// lookahead is not a grammar sym
let errmsg = if let Some(Error(em)) = &actionopt {
format!("unexpected symbol {} on line {}, column {}: ** {} ** ..",lksym,lookahead.line,lookahead.column,em.trim())
} else {format!("unexpected symbol {} on line {}, column {} .. ",lksym,lookahead.line,lookahead.column)};
parser.report(&errmsg);
if self.training { ////// Training mode
let csym = lookahead.sym.to_owned();
let mut inp = String::from("");
if let None=self.scriptinopt { // interactive mode
if let Some(outfd1) = &self.scriptoutopt {
let mut outfd = outfd1;
print!("\n>>>TRAINER: if this message is not adequate (for state {}), enter a replacement (default no change): ",cstate);
let rrrflush = io::stdout().flush();
if let Ok(n) = io::stdin().read_line(&mut inp) {
if inp.len()>5 && parser.Symset.contains(lksym) {
print!(">>>TRAINER: should this message be given for all unexpected symbols in the current state? (default yes) ");
let rrrflush2 = io::stdout().flush();
let mut inp2 = String::new();
if let Ok(n) = io::stdin().read_line(&mut inp2) {
if inp2.trim()=="no" || inp2.trim()=="No" {
wresult = write!(outfd,"{}\t{}\t{} ::: {}\n",parser.linenum,parser.column,&csym,inp.trim());
self.trained.insert((cstate,csym),inp);
}
else {// insert for any error
wresult = write!(outfd,"{}\t{}\t{} ::: {}\n",parser.linenum,parser.column,"ANY_ERROR",inp.trim());
self.trained.insert((cstate,String::from("ANY_ERROR")),inp);
}
}// read ok
}// unexpected symbol is grammar sym
else if inp.len()>5 && !parser.Symset.contains(lksym) {
wresult = write!(outfd,"{}\t{}\t{} ::: {}\n",parser.linenum,parser.column,"ANY_ERROR",inp.trim());
self.trained.insert((cstate,String::from("ANY_ERROR")),inp);
}
}// process user response
}}// interactive mode
else { // training from script mode (non-interactive)
if let Some(brfd) = &mut self.scriptinopt {
let mut scin = brfd;
let mut readn = 0;
while readn < 1
{
inp = String::new();
match scin.read_line(&mut inp) {
Ok(n) if n>1 && &inp[0..1]!="#" && inp.trim().len()>0 => {readn=n;},
Ok(n) if n>0 => { readn=0; }, // keep reading
_ => {readn = 1; } // stop - this means End of Stream
}//match
if readn>1 { // read something
let inpsplit:Vec<&str> = inp.split_whitespace().collect();
if inpsplit.len()>4 && inpsplit[3].trim()==":::" {
let inline = inpsplit[0].trim().parse::<usize>().unwrap();
let incolumn = inpsplit[1].trim().parse::<usize>().unwrap();
let insym = inpsplit[2].trim();
if parser.linenum==inline && parser.column==incolumn {
if &csym==insym || insym=="ANY_ERROR" {
let posc = inp.find(":::").unwrap()+4;
println!("\n>>>Found matching entry from training script for {}, error message: {}",insym,&inp[posc..]);
self.trained.insert((cstate,String::from(insym)),String::from(&inp[posc..]));
} // unexpected symbol match
}// line/column match
}//inpsplit check
}// valid training line read
}//while readn<2
}}//training from script mode
}//if training //// END TRAINING MODE
}// standardreporter function
}// impl ErrReporter for StandardReporter
/////////////////////////////////////////////////////////////
//////////////// parse_core replaced: now uses zc tokenizer
impl<AT:Default,ET:Default> ZCParser<AT,ET>
{
pub fn parse_core<'u,'t:'u>(&mut self, tokenizer:&'u mut dyn Tokenizer<'t,AT>, err_handler:&mut dyn ErrReporter<AT,ET>) -> AT
{
self.stack.clear();
self.err_occurred = false;
let mut result = AT::default();
self.stack.push(StackedItem::new(0,AT::default(),0,0));
//self.stack.push(Stackelement {si:0, value:AT::default()});
self.stopparsing = false;
let mut action = Stateaction::Error("");
let mut lookahead = TerminalToken::new("EOF",AT::default(),0,0); //just init
// nextsym() should only be called here
if let Some(tok) = tokenizer.nextsym() {lookahead=tok;}
else {self.stopparsing=true;}
while !self.stopparsing
{
self.linenum = self.stack[self.stack.len()-1].line;
self.column=self.stack[self.stack.len()-1].column;
let currentstate = self.stack[self.stack.len()-1].si;
let mut actionopt = self.RSM[currentstate].get(lookahead.sym);
let actclone:Option<Stateaction> = match actionopt {
Some(a) => Some(*a),
None => None,
};
if iserror(&actionopt) { // either None or Error
if !self.err_occurred {self.err_occurred = true;}
err_handler.err_reporter(self,&lookahead,&actclone);
//err_reporter(self,&lookahead,&actclone);
match self.error_recover(&mut lookahead,tokenizer) {
None => { self.stopparsing=true; break; }
Some(act) => {action = act;}, // lookahead=la;},
}//match
}// iserror
else { action = actclone.unwrap(); }
match &action {
Shift(nextstate) => {
lookahead = self.shift(*nextstate,lookahead,tokenizer);
},
Reduce(rulei) => { self.reduce(rulei); },
Accept => {
self.stopparsing=true;
if self.stack.len()>0 {result = self.stack.pop().unwrap().value;}
else {self.err_occurred=true;}
},
_ => {}, // continue
}//match action
}// main parse loop
return result;
}//parse_core
///provided generic parsing function that reports errors on std::io.
pub fn parse<'t>(&mut self, tokenizer:&mut dyn Tokenizer<'t,AT>) -> AT
{
let mut stdeh = StandardReporter::new();
self.parse_core(tokenizer,&mut stdeh)
}//parse_stdio
///Parses in interactive training mode with provided path to parserfile.
///The parser file will be modified and a training script file will be
///created for future retraining after grammar is modified.
///
/// When an error occurs, the parser will
/// ask the human trainer for an appropriate error message: it will
/// then insert an entry into its state transition table to
/// give the same error message on future errors of the same type.
/// If the error is caused by an unexpected token that is recognized
/// as a terminal symbol of the grammar, the trainer can select to
/// enter the entry
/// under the reserved ANY_ERROR symbol. If the unexpected token is
/// not recognized as a grammar symbol, then the entry will always
/// be entered under ANY_ERROR. ANY_ERROR entries for a state will match
/// all future unexpected symbols for that state: however, entries for
/// valid grammar symbols will still override the generic entry.
///
/// Example: with the parser for this [toy grammar](https://cs.hofstra.edu/~cscccl/rustlr_project/cpm.grammar), parse_train can run as follows:
///```ignore
/// Write something in C+- : cout << x y ;
/// ERROR on line 1, column 0: unexpected symbol y ..
/// >>>TRAINER: is this error message adequate? If not, enter a better one: need another <<
/// >>>TRAINER: should this message be given for all unexpected symbols in the current state? (default yes) yes
///```
/// (ignore the column number as the lexer for this toy language does not implement it)
///
/// parse_train will then produced a [modified parser](https://cs.hofstra.edu/~cscccl/rustlr_project/cpmparser.rs) as specified
/// by the filename (path) argument. When the augmented parser is used, it will
/// give a more helpful error message:
///```
/// Write something in C+- : cout << x endl
/// ERROR on line 1, column 0: unexpected symbol endl, ** need another << ** ..
///```
///
/// parse_stdio_train calls parse_stdio, which uses stdin/stdout for user interface.
/// Parsing in interactive training mode also produces a [training script file](http://cs.hofstra.edu/~cscccl/rustlr_project/cpmparser.rs_script.txt) which can
/// be used to re-train a parser using [ZCParser::train_from_script].
/// This is useful after a grammar is modified with extensions to a language.
pub fn parse_train<'t>(&mut self, tokenizer:&mut dyn Tokenizer<'t,AT>, parserfile:&str) -> AT
{
let mut stdtrainer = StandardReporter::new_interactive_training(parserfile);
let result = self.parse_core(tokenizer,&mut stdtrainer);
if let Err(m) = stdtrainer.augment_training(parserfile) {
eprintln!("Error in augmenting parser: {:?}",m)
}
return result;
}//parse_stdio_train
/// trains parser from a [training script](https://cs.hofstra.edu/~cscccl/rustlr_project/cpmparser.rs_script.txt)
/// created by interactive training. This
/// is intended to be used after a grammar has been modified and the parser
/// is regenerated with different state numbers. It is the user's
/// responsibility to keep consistent the parser file, script file, and sample
/// input that was used when the script was created. The script contains
/// the line and column numbers of each error encountered, along with either
/// the unexpected symbol that caused the error, or the reserved ANY_ERROR
/// symbol if the error message is to be applied to all unexpected symbols.
/// These entries must match, in sequence, the errors encountered during
/// retraining - it is therefore recommended that the same tokenizer be used
/// during retraining so that the same line/column information are given.
/// The trainer will augment the parser (parserfile) with new Error
/// entries, overriding any previous ones. It is also recommended that the
/// user examines the "load_extras" function that appears at the end of
/// the [augmented parser](https://cs.hofstra.edu/~cscccl/rustlr_project/cpmparser.rs).
/// The train_from_script function does not return
/// a value, unlike [ZCParser::parse] and [ZCParser::parse_train].
pub fn train_from_script<'t>(&mut self, tokenizer:&mut dyn Tokenizer<'t,AT>,parserfile:&str, scriptfile:&str)
{
let mut stdtrainer = StandardReporter::new_script_training(parserfile,scriptfile);
let result = self.parse_core(tokenizer,&mut stdtrainer);
if let Err(m) = stdtrainer.augment_training(parserfile) {
eprintln!("Error in augmenting parser: {:?}",m)
}
if !self.err_occurred {println!("no errors encountered during parsing");}
}//train_from_script
}// 3rd impl ZCParser