path: root/src/intertyper.js

//"use strict";

// LLVM assembly => internal intermediate representation, which is ready
// to be processed by the later stages.

var tokenizer; // TODO: Clean this up/out
               //       XXX In particular, this closes over the substrate, which can keep stuff in memory, which is bad
function tokenize(text) {
  return tokenizer.processItem({ lineText: text }, true);
}

// Handy sets

var ENCLOSER_STARTERS = set('[', '(', '<');
var ENCLOSER_ENDERS = {
  '[': ']',
  '(': ')',
  '<': '>'
};
var ZEROINIT_UNDEF = set('zeroinitializer', 'undef');
var NSW_NUW = set('nsw', 'nuw');

// Intertyper

function intertyper(data, sidePass, baseLineNums) {
  var mainPass = !sidePass;
  baseLineNums = baseLineNums || [[0,0]]; // each pair [#0,#1] means "starting from line #0, the base line num is #1"

  dprint('framework', 'Big picture: Starting intertyper, main pass=' + mainPass);

  // Substrate

  var substrate = new Substrate('Intertyper');

  // Line splitter. We break off some bunches of lines into unparsedBundles, which are
  // parsed in separate passes later. This helps to keep memory usage low - we can start
  // from raw lines and end up with final JS for each function individually that way, instead
  // of intertyping them all, then analyzing them all, etc.
  substrate.addActor('LineSplitter', {
    processItem: function _lineSplitter(item) {
      var lines = item.llvmLines;
      var ret = [];
      var inContinual = false;
      var inFunction = false;
      var currFunctionLines;
      var currFunctionLineNum;
      var unparsedBundles = [];
      var unparsedTypes, unparsedGlobals;
      if (mainPass) {
        unparsedTypes = {
          intertype: 'unparsedTypes',
          lines: []
        };
        unparsedBundles.push(unparsedTypes);
        unparsedGlobals = {
          intertype: 'unparsedGlobals',
          lines: []
        };
        unparsedBundles.push(unparsedGlobals);
      }
      var baseLineNumPosition = 0;
      for (var i = 0; i < lines.length; i++) {
        var line = lines[i];
        if (singlePhase) lines[i] = null; // lines may be very very large. Allow GCing to occur in the loop by releasing refs here

        while (baseLineNumPosition < baseLineNums.length-1 && i >= baseLineNums[baseLineNumPosition+1][0]) {
          baseLineNumPosition++;
        }

        if (mainPass && (line[0] == '%' || line[0] == '@')) {
          // If this isn't a type, it's a global variable, make a note of the information now, we will need it later
          var parts = line.split(' = ');
          assert(parts.length >= 2);
          var left = parts[0], right = parts.slice(1).join(' = ');
          var testType = /^type .*/.exec(right);
          if (!testType) {
            var globalIdent = toNiceIdent(left);
            var testAlias = /^(hidden )?alias .*/.exec(right);
            Variables.globals[globalIdent] = {
              name: globalIdent,
              alias: !!testAlias,
              impl: VAR_EMULATED
            };
            unparsedGlobals.lines.push(line);
          } else {
            unparsedTypes.lines.push(line);
          }
          continue;
        }
        if (mainPass && /^define .*/.test(line)) {
          inFunction = true;
          currFunctionLines = [];
          currFunctionLineNum = i + 1;
        }
        if (!inFunction || !mainPass) {
          if (inContinual || /^\ +(to|catch |filter |cleanup).*/.test(line)) {
            // to after invoke or landingpad second line
            ret.slice(-1)[0].lineText += line;
            if (/^\ +\]/.test(line)) { // end of llvm switch
              inContinual = false;
            }
          } else {
            ret.push({
              lineText: line,
              lineNum: i + 1 + baseLineNums[baseLineNumPosition][