schala/schala-lang/language/src/schala.rs

use stopwatch::Stopwatch;

use std::cell::RefCell;
use std::rc::Rc;
use std::collections::HashSet;

use itertools::Itertools;
use schala_repl::{ProgrammingLanguageInterface,
ComputationRequest, ComputationResponse,
LangMetaRequest, LangMetaResponse, GlobalOutputStats,
DebugResponse, DebugAsk};
use crate::{ast, reduced_ast, tokenizing, parsing, eval, typechecking, symbol_table, source_map};

pub type SymbolTableHandle = Rc<RefCell<symbol_table::SymbolTable>>;
pub type SourceMapHandle = Rc<RefCell<source_map::SourceMap>>;

/// All the state necessary to parse and execute a Schala program are stored in this struct.
/// `state` represents the execution state for the AST-walking interpreter, the other fields
/// should be self-explanatory.
#[allow(dead_code)]
pub struct Schala {
  source_reference: SourceReference,
  source_map: SourceMapHandle,
  state: eval::State<'static>,
  symbol_table: SymbolTableHandle,
  resolver: crate::scope_resolution::ScopeResolver<'static>,
  type_context: typechecking::TypeContext<'static>,
  active_parser: parsing::Parser,
}

impl Schala {
  //TODO implement documentation for language items
  /*
  fn handle_docs(&self, source: String) -> LangMetaResponse {
    LangMetaResponse::Docs {
      doc_string: format!("Schala item `{}` : <<Schala-lang documentation not yet implemented>>", source)
    }
  }
  */
}

impl Schala {
  /// Creates a new Schala environment *without* any prelude.
  fn new_blank_env() -> Schala {
    let source_map = Rc::new(RefCell::new(source_map::SourceMap::new()));
    let symbols = Rc::new(RefCell::new(symbol_table::SymbolTable::new(source_map.clone())));
    Schala {
      //TODO maybe these can be the same structure
      source_reference: SourceReference::new(),
      symbol_table: symbols.clone(),
      source_map: source_map.clone(),
      resolver: crate::scope_resolution::ScopeResolver::new(symbols.clone()),
      state: eval::State::new(),
      type_context: typechecking::TypeContext::new(),
      active_parser: parsing::Parser::new(source_map)
    }
  }

  /// Creates a new Schala environment with the standard prelude, which is defined as ordinary
  /// Schala code in the file `prelude.schala`
  pub fn new() -> Schala {
    let prelude = include_str!("prelude.schala");
    let mut s = Schala::new_blank_env();

    //TODO this shouldn't depend on schala-repl types
    let request = ComputationRequest { source: prelude, debug_requests: HashSet::default() };
    let response = s.run_computation(request);
    if let Err(msg) = response.main_output {
      panic!("Error in prelude, panicking: {}", msg);
    }
    s
  }

  /// This is where the actual action of interpreting/compilation happens.
  /// Note: this should eventually use a query-based system for parallelization, cf.
  /// https://rustc-dev-guide.rust-lang.org/overview.html
  fn run_pipeline(&mut self, source: &str) -> Result<String, String> {
    //TODO `run_pipeline` should return a formatted error struct

    //TODO every stage should have a common error-format that gets turned into a repl-appropriate
    //string in `run_computation`
    // 1st stage - tokenization
    // TODO tokenize should return its own error type
    let tokens = tokenizing::tokenize(source);
    let token_errors: Vec<String> = tokens.iter().filter_map(|t| t.get_error()).collect();
    if token_errors.len() > 0 {
      return Err(format!("{:?}", token_errors));
    }

    //2nd stage - parsing
    self.active_parser.add_new_tokens(tokens);
    let mut ast = self.active_parser.parse()
      .map_err(|err| format_parse_error(err, &self.source_reference))?;

    // Symbol table
    self.symbol_table.borrow_mut().add_top_level_symbols(&ast)?;

    // Scope resolution - requires mutating AST
    self.resolver.resolve(&mut ast)?;

    // Typechecking
    let overall_type = self.type_context.typecheck(&ast)
      .map_err(|err| format!("Type error: {}", err.msg))?;

    // Reduce AST - this doesn't produce an error yet, but probably should
    let symbol_table = self.symbol_table.borrow();
    let reduced_ast = reduced_ast::reduce(&ast, &symbol_table);

    // Tree-walking evaluator. TODO fix this
    let evaluation_outputs = self.state.evaluate(reduced_ast, true);
    let text_output: Result<Vec<String>, String> = evaluation_outputs
      .into_iter()
      .collect();

    let eval_output: String = text_output
      .map(|v| { Iterator::intersperse(v.into_iter(), "\n".to_owned()).collect() })?;

      Ok(eval_output)
  }
}

fn parsing(input: Vec<tokenizing::Token>, handle: &mut Schala, comp: Option<&mut PassDebugArtifact>) -> Result<ast::AST, String> {
  use ParsingDebugType::*;

  let ref mut parser = handle.active_parser;
  parser.add_new_tokens(input);
  let ast = parser.parse();

  comp.map(|comp| {
    let debug_format = comp.parsing.as_ref().unwrap_or(&CompactAST);
    let debug_info = match debug_format {
      CompactAST => match ast{
        Ok(ref ast) => ast.compact_debug(),
        Err(_) => "Error - see output".to_string(),
      },
      ExpandedAST => match ast{
        Ok(ref ast) => ast.expanded_debug(),
        Err(_) => "Error - see output".to_string(),
      },
      Trace => parser.format_parse_trace(),
    };
    comp.add_artifact(debug_info);
  });
  ast.map_err(|err| format_parse_error(err, &handle.source_reference))
}

fn format_parse_error(error: parsing::ParseError, source_reference: &SourceReference) -> String {
  let line_num = error.token.location.line_num;
  let ch = error.token.location.char_num;
  let line_from_program = source_reference.get_line(line_num);
  let location_pointer = format!("{}^", " ".repeat(ch));

  let line_num_digits = format!("{}", line_num).chars().count();
  let space_padding = " ".repeat(line_num_digits);

  let production = match error.production_name {
    Some(n) => format!("\n(from production \"{}\")", n),
    None => "".to_string()
  };

  format!(r#"
{error_msg}{production}
{space_padding} |
{line_num} | {}
{space_padding} | {}
"#, line_from_program, location_pointer, error_msg=error.msg, space_padding=space_padding, line_num=line_num, production=production
)
}

fn symbol_table(input: ast::AST, handle: &mut Schala, comp: Option<&mut PassDebugArtifact>) -> Result<ast::AST, String> {
  let () = handle.symbol_table.borrow_mut().add_top_level_symbols(&input)?;
  comp.map(|comp| {
    let debug = handle.symbol_table.borrow().debug_symbol_table();
    comp.add_artifact(debug);
  });
  Ok(input)
}

fn scope_resolution(mut input: ast::AST, handle: &mut Schala, _com: Option<&mut PassDebugArtifact>) -> Result<ast::AST, String> {
  let () = handle.resolver.resolve(&mut input)?;
  Ok(input)
}

fn typechecking(input: ast::AST, handle: &mut Schala, comp: Option<&mut PassDebugArtifact>) -> Result<ast::AST, String> {
  let result = handle.type_context.typecheck(&input);

  comp.map(|comp| {
    comp.add_artifact(match result {
      Ok(ty) => ty.to_string(),
      Err(err) => format!("Type error: {}", err.msg)
    });
  });

  Ok(input)
}

fn ast_reducing(input: ast::AST, handle: &mut Schala, comp: Option<&mut PassDebugArtifact>) -> Result<reduced_ast::ReducedAST, String> {
  let ref symbol_table = handle.symbol_table.borrow();
  let output = reduced_ast::reduce(&input, symbol_table);
  comp.map(|comp| comp.add_artifact(format!("{:?}", output)));
  Ok(output)
}

fn eval(input: reduced_ast::ReducedAST, handle: &mut Schala, comp: Option<&mut PassDebugArtifact>) -> Result<String, String> {
  comp.map(|comp| comp.add_artifact(handle.state.debug_print()));
  let evaluation_outputs = handle.state.evaluate(input, true);
  let text_output: Result<Vec<String>, String> = evaluation_outputs
    .into_iter()
    .collect();

  let eval_output: Result<String, String> = text_output
    .map(|v| { Iterator::intersperse(v.into_iter(), "\n".to_owned()).collect() });
  eval_output
}

/// Represents lines of source code
struct SourceReference {
  lines: Option<Vec<String>>
}

impl SourceReference {
  fn new() -> SourceReference {
    SourceReference { lines: None }
  }

  fn load_new_source(&mut self, source: &str) {
    //TODO this is a lot of heap allocations - maybe there's a way to make it more efficient?
    self.lines = Some(source.lines().map(|s| s.to_string()).collect()); }

  fn get_line(&self, line: usize) -> String {
    self.lines.as_ref().and_then(|x| x.get(line).map(|s| s.to_string())).unwrap_or(format!("NO LINE FOUND"))
  }
}

enum ParsingDebugType {
  CompactAST,
  ExpandedAST,
  Trace
}

#[derive(Default)]
struct PassDebugArtifact {
  parsing: Option<ParsingDebugType>,
  artifacts: Vec<String>

}
impl PassDebugArtifact {
  fn add_artifact(&mut self, artifact: String) {
    self.artifacts.push(artifact)
  }
}

fn stage_names() -> Vec<&'static str> {
  vec![
    "tokenizing",
    "parsing",
    "symbol-table",
    "scope-resolution",
    "typechecking",
    "ast-reduction",
    "ast-walking-evaluation"
  ]
}


impl ProgrammingLanguageInterface for Schala {
  fn language_name(&self) -> String {
    "Schala".to_owned()
  }

  fn source_file_suffix(&self) -> String {
    "schala".to_owned()
  }

  fn run_computation(&mut self, request: ComputationRequest) -> ComputationResponse {
    let ComputationRequest { source, debug_requests } = request;
    self.source_reference.load_new_source(source);
    let sw = Stopwatch::start_new();

    let main_output = self.run_pipeline(source);

    let global_output_stats = GlobalOutputStats {
      total_duration: sw.elapsed(),
      stage_durations: vec![]
    };

    ComputationResponse {
      main_output,
      global_output_stats,
      debug_responses: vec![]
    }
  }

  // n.b. - old-style, overcomplicated `run_computation`. need to revamp the entire metadata system
  /*
  fn run_computation(&mut self, request: ComputationRequest) -> ComputationResponse {
    struct PassToken<'a> {
      schala: &'a mut Schala,
      stage_durations: &'a mut Vec<(String, Duration)>,
      sw: &'a Stopwatch,
      debug_requests: &'a HashSet<DebugAsk>,
      debug_responses: &'a mut Vec<DebugResponse>,
    }

    fn output_wrapper<Input, Output, F>(n: usize, func: F, input: Input, token: &mut PassToken) -> Result<Output, String>
      where F: Fn(Input, &mut Schala, Option<&mut PassDebugArtifact>) -> Result<Output, String>
    {
      let stage_names = stage_names();
      let cur_stage_name = stage_names[n];
      let ask = token.debug_requests.iter().find(|ask| ask.is_for_stage(cur_stage_name));

      let parsing = match ask {
        Some(DebugAsk::ByStage { token, .. }) if cur_stage_name == "parsing" => Some(
          token.as_ref().map(|token| match &token[..] {
            "compact" => ParsingDebugType::CompactAST,
            "expanded" => ParsingDebugType::ExpandedAST,
            "trace" => ParsingDebugType::Trace,
            _ => ParsingDebugType::CompactAST,
          }).unwrap_or(ParsingDebugType::CompactAST)
        ),
        _ => None,
      };

      let mut debug_artifact = ask.map(|_| PassDebugArtifact {
        parsing, ..Default::default()
      });

      let output = func(input, token.schala, debug_artifact.as_mut());

      //TODO I think this is not counting the time since the *previous* stage
      token.stage_durations.push((cur_stage_name.to_string(), token.sw.elapsed()));
      if let Some(artifact) = debug_artifact {
        for value in artifact.artifacts.into_iter() {
          let resp = DebugResponse { ask: ask.unwrap().clone(), value };
          token.debug_responses.push(resp);
        }
      }
      output
    }

    let ComputationRequest { source, debug_requests } = request;
    self.source_reference.load_new_source(source);
    let sw = Stopwatch::start_new();
    let mut stage_durations = Vec::new();
    let mut debug_responses = Vec::new();
    let mut tok = PassToken { schala: self, stage_durations: &mut stage_durations, sw: &sw, debug_requests: &debug_requests, debug_responses: &mut debug_responses };

    let main_output: Result<String, String> = Ok(source)
      .and_then(|source| output_wrapper(0, tokenizing, source, &mut tok))
      .and_then(|tokens| output_wrapper(1, parsing, tokens, &mut tok))
      .and_then(|ast| output_wrapper(2, symbol_table, ast, &mut tok))
      .and_then(|ast| output_wrapper(3, scope_resolution, ast, &mut tok))
      .and_then(|ast| output_wrapper(4, typechecking, ast, &mut tok))
      .and_then(|ast| output_wrapper(5, ast_reducing, ast, &mut tok))
      .and_then(|reduced_ast| output_wrapper(6, eval, reduced_ast, &mut tok));

    let total_duration = sw.elapsed();
    let global_output_stats = GlobalOutputStats {
      total_duration, stage_durations
    };

    ComputationResponse {
      main_output,
      global_output_stats,
      debug_responses,
    }
  }
  */

  fn request_meta(&mut self, request: LangMetaRequest) -> LangMetaResponse {
    match request {
      LangMetaRequest::StageNames => LangMetaResponse::StageNames(stage_names().iter().map(|s| s.to_string()).collect()),
      _ => LangMetaResponse::Custom { kind: format!("not-implemented"), value: format!("") }
    }
  }
}