2011-12-14

 * ./tomrec.xhtml & ./tomrec_web.js: forbade the use of `arguments` within a function to be unrolled. The error appears automatically in the interactive area after one pressed the "eliminate tail calls" button.

    * Implementation: the call `tt.check_tree_no_arguments( top_ftree );` in ./tool.tailopt.unroll.js

    * Unit test: ./test.unit.tailopt.no_arguments.js

    * Rationale:

            //       http://wiki.ecmascript.org/doku.php?id=strawman:proper_tail_calls
            //       " Since ES5/strict and ES-Harmony
            //       "
            //       "     * poisons <function>.caller and <function>.arguments,
            //       "     * prohibits <non-strict-function>.caller from revealing a strict function,
            //       "     * poisons arguments.callee and arguments.caller, and
            //       "     * does not join arguments with parameters
            //       "
            //       " all calls in possible tail position MAY safely be implemented as PTC by our gc semantics safety rule. 

2011-12-13
 
 * ./tomrec.xhtml & ./tomrec_web.js: the interactive area now reports syntax error and tailopt errors after one pressed the "eliminate tail calls" button.

2011-12-09

 * Moved to Python 2.7

 * ./tool.tailopt.unroll.js - Added an error detection when trying to tail-optimize "impure" code, i.e. code that mixes tail calls and non-tail calls. This should help people to get a clearer idea of what tail calls are.

    * Example of impure code:

        function no_tail_sum(pair) {
            return _sum( 0, pair );      // <<<<< tail call (OK)
            function _sum( acc, x ) {
                if (typeof x === 'object'  &&  2 === x.length)
                {
                    acc = _sum( acc, x[0] );   // <<<<< non-tail call on the function `_sum` that `tailopt` is trying to unwrap -> meaningless result
                    acc = _sum( acc, x[1] );   // <<<<< non-tail call on the function `_sum` that `tailopt` is trying to unwrap -> meaningless result
                    return acc;
                }

                return acc + x;
            }
        }
        no_tail_sum([[1,2],[3,[[4,5],6]]])
        // >>> 21

    * For more "impure" examples, see the unit test: ./test.unit.tool.tailopt.not_a_tailcall.js

    * Implementation details: look for: checkthrow_all_calls_removed in ./tool.tailopt.unroll.js
        // Throws an Error if (1) the code `ret` has a syntax error or
        // (2) not all calls have been removed.
        // 
        // (2) makes sure all calls to dependencies have been removed,
        // i.e. makes sure that the original code does NOT contain any
        // non-tail call to a function it *also* tail calls.
        //
        // Test added after receiving feedback from Nikita Vasilev on
        // 2011-12-09

    * Added a safety to test.unit.js to make sure all tests are properly run.
       * Details: look for:   auto_test_arr.sort();

2011-08-19

 * Fixed bug concerning the translation of variable names in a specific "homonyms" case.
 * Added unit test for it.
 * Details: The bug was that this:
       function homonyms (a, b) { return sub(a, b); function sub(a, b) { return { a : a, b : b }; } }

   was mistakenly tail-optimized into this:
       function homonyms(a,b)
       {
           var undef,a,b,sub,a0,b0;
           a0=a; b0=b;
           _L_sub_: while(true)
           {
               return {a0:a0,b0:b0};   // <<< Bug: The object keys should have been `a` and `b`.
               return;
           }
        }

   Added the "homonyms" unit test case in ./test.unit.tailopt_str.js,
   then fixed the bug in `ttu.mapper_gen` in ./tool.tailopt.unroll.js

   Now the above example is correctly tail-optimized into this:
       function homonyms(a,b)
       {
           var undef,a,b,sub,a0,b0;
           a0=a; b0=b;
           _L_sub_: while(true)
           {
               return {a:a0,b:b0};   // <<< Fixed: The object keys are correct: `a` and `b`.
               return;
           }
        }

...

2010-11-06

 * narcissus.jsparse.js: added the string constant
   narcissus.jsparse.MISSING_FUNCTION_IDENTIFIER

2010-06-17

 * narcissus.jsdef.js: dropped the eval in the const generation 
   to be able to pass narcissus.jsdef.js through a build system.

2010-04-21

 * narcissus.jsparse.js: added from.req() safety to make sure
   narcissus.jsdef.js is fully interpreted first. The code should
   still work without from.js .

2010-04-20

 * narcissus.jsparse.js: removed "callable RegExp object" SpiderMonkey
   extension, replacing:
        } else if ((match = opRegExp(input))) {
   with:
        } else if ((match = opRegExp.exec(input))) {

2010-04-14

 * jscheck_web.js: added Opera workaround for textarea editing.

 * narcissus.jsparse.js: removed Array.splice SpiderMonkey extension.
   This line:
        var a = operands.splice(operands.length - arity);
   was replaced with:
        var a = operands.splice(operands.length - arity, arity);


2010-04-01

 * narcissus.jsparse.js and narcissus.jsdef.js: fixed small details
   (extra comma, missing var) so that both now pass JSCheck without
   warning.

2010-03-31

 * narcissus.jsparse.js: Added detection of empty child expressions cases, 
   such as [ 'a', , 'b', 'c' ]

// Always use push to add operands to an expression, to update start and end.
Np.push = function (kid) {
>>> The next two lines were inserted
    if (!kid) 
        throw this.tokenizer.newSyntaxError('Empty child expression!');
<<< End of insertion
    if (kid.start < this.start)
        this.start = kid.start;
    if (this.end < kid.end)
        this.end = kid.end;
    return Array.prototype.push.call(this, kid);
}

2010-03-04

 * jsdef.js & jsparse.js: Fixed unreliable order bug "for (var a in opTypeNames)"
   introducing opTypeNamesArr. Now opRegExpSrc works in Rhino 1.6 as well.

 * jsdef.js & jsparse.js: In keywords, removed __proto__ spidermonkey extension,
   replaced it with hasOwnProperty. Now keywords(id) works in Rhino 1.6 as well.

2010-02-19

 * Wrote a.html & a.js test case
 * Got Narcissus to work on Firefox by replacing the three __defineProperty__ / __defineGetter__ in jsparse.js
 * Fixed comment detection issue in line 140 of jsparse.js, replacing:
               if (!(match = /^\/(?:\*(?:.|\n)*?\*\/|\/.*)/(input)))
   with:
               if (!(match = /^\/(?:\*(?:[\S\s])*?\*\/|\/.*)/(input)))
 * First successful run of the test case a.html

2010-02-19

 * With FF 3.5.8 added a truth test with out.toSource