path: root/src/parseTools.js

// Various tools for parsing LLVM. Utilities of various sorts, that are
// specific to Emscripten (and hence not in utility.js).

// Does simple 'macro' substitution, using Django-like syntax,
// {{{ code }}} will be replaced with |eval(code)|.
function processMacros(text) {
  return text.replace(/{{{[^}]+}}}/g, function(str) {
    str = str.substr(3, str.length-6);
    return eval(str).toString();
  });
}

// Simple #if/else/endif preprocessing for a file. Checks if the
// ident checked is true in our global. Also replaces some constants.
function preprocess(text, constants) {
  for (constant in constants) {
    text = text.replace(eval('/' + constant + '/g'), constants[constant]);
  }
  var lines = text.split('\n');
  var ret = '';
  var showStack = [];
  for (var i = 0; i < lines.length; i++) {
    var line = lines[i];
    if (!line[0] || line[0] != '#') {
      if (showStack.indexOf(false) == -1) {
        ret += line + '\n';
      }
    } else {
      if (line[1] && line[1] == 'i') { // if
        var ident = line.substr(4);
        showStack.push(!!this[ident] && this[ident] > 0);
      } else if (line[2] && line[2] == 'l') { // else
        showStack.push(!showStack.pop());
      } else if (line[2] && line[2] == 'n') { // endif
        showStack.pop();
      } else {
        throw "Unclear preprocessor command: " + line;
      }
    }
  }
  assert(showStack.length == 0);
  return ret;
}

function addPointing(type) { return type + '*' }
function removePointing(type, num) {
  if (num === 0) return type;
  return type.substr(0, type.length-(num ? num : 1));
}

function pointingLevels(type) {
  if (!type) return 0;
  var ret = 0;
  var len1 = type.length - 1;
  while (type[len1-ret] && type[len1-ret] === '*') {
    ret ++;
  }
  return ret;
}

function removeAllPointing(type) {
  return removePointing(type, pointingLevels(type));
}

function toNiceIdent(ident) {
  assert(ident);
  if (parseFloat(ident) == ident) return ident;
  if (ident == 'null') return '0'; // see parseNumerical
  return ident.replace('%', '$').replace(/["&\\ \.@:<>,\*\[\]-]/g, '_');
}

// Kind of a hack. In some cases we have strings that we do not want
// to |toNiceIdent|, as they are the output of previous processing. We
// should refactor everything into an object, with an explicit flag
// saying what has been |toNiceIdent|ed. Until then, this will detect
// simple idents that are in need of |toNiceIdent|ation. Or, we should
// ensure that processed strings never start with %,@, e.g. by always
// enclosing them in ().
function toNiceIdentCarefully(ident) {
  if (ident[0] == '%' || ident[0] == '@') ident = toNiceIdent(ident);
  return ident;
}

function isStructPointerType(type) {
  // This test is necessary for clang - in llvm-gcc, we
  // could check for %struct. The downside is that %1 can
  // be either a variable or a structure, and we guess it is
  // a struct, which can lead to |call i32 %5()| having
  // |%5()| as a function call (like |i32 (i8*)| etc.). So
  // we must check later on, in call(), where we have more
  // context, to differentiate such cases.
  // A similar thing happns in isStructType()
  return !Runtime.isNumberType(type) && type[0] == '%';
}

function isStructType(type) {
  if (isPointerType(type)) return false;
  if (new RegExp(/^\[\d+\ x\ (.*)\]/g).test(type)) return true; // [15 x ?] blocks. Like structs
  // See comment in isStructPointerType()
  return !Runtime.isNumberType(type) && type[0