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 MATHOPS = set(['add', 'sub', 'sdiv', 'udiv', 'mul', 'icmp', 'zext', 'urem', 'srem', 'fadd', 'fsub', 'fmul', 'fdiv', 'fcmp', 'uitofp', 'sitofp', 'fpext', 'fptrunc', 'fptoui', 'fptosi', 'trunc', 'sext', 'select', 'shl', 'shr', 'ashl', 'ashr', 'lshr', 'lshl', 'xor', 'or', 'and', 'ptrtoint', 'inttoptr']);
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);

  if (mainPass) {
    if (LLVM_STYLE === null) {
      // new = clang on 2.8, old = llvm-gcc anywhere or clang on 2.7
      LLVM_STYLE = (data.indexOf('<label>') == -1 && data.indexOf('entry:') != -1) ? 'old' : 'new';
      //dprint('LLVM_STYLE: ' + LLVM_STYLE);
    }
  }

  // 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];
        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 testType = /[@%\w\d\.\" ]+ = type .*/.exec(line);
          if (!testType) {
            var global = /([@%\w\d\.\" ]+) = .*/.exec(line);
            var globalIdent = toNiceIdent(global[1]);
            var testAlias = /[@%\w\d\.\" ]+ = alias .*/.exec(line);
            var testString = /^[^"]+c\"[^"]+"/.exec(line);
            Variables.globals[globalIdent] = {
              name: globalIdent,
              alias: !!testAlias,
              impl: VAR_EMULATED,
              isString : !!testString
            };
            unparsedGlobals.lines.push(line);
          } else {
            unparsedTypes.lines.push(line);
          }
          continue;
        }
        if (mainPass && /^define .*/.test(line