Remove completion value.

[zherczeg]

Remove ecma_completion_value_t, and an extra bit to ecma_value_t to represent errors. From the long list of completion types only normal and error remained.

[bzsolt]

RPi2 performance measurements
master 977bfa5

                     Benchmark |              Perf<br>(+ is better)
                    3d-cube.js |  2.686s ->   2.644s :  +1.564%  (+-0.589%) : [+]
        access-binary-trees.js |  1.422s ->   1.408s :  +0.985%  (+-1.705%) : [≈]
            access-fannkuch.js |  8.026s ->   7.954s :  +0.897%  (+-1.187%) : [≈]
               access-nbody.js |  3.158s ->   3.098s :  +1.900%  (+-1.001%) : [+]
   bitops-3bit-bits-in-byte.js |  1.784s ->   1.758s :  +1.457%  (+-0.809%) : [+]
        bitops-bits-in-byte.js |  2.748s ->   2.718s :  +1.092%  (+-1.029%) : [+]
         bitops-bitwise-and.js |  3.788s ->   3.718s :  +1.848%  (+-1.067%) : [+]
      controlflow-recursive.js |  0.946s ->   0.896s :  +5.285%  (+-1.642%) : [+]
                 crypto-aes.js |  4.640s ->   4.658s :  -0.388%  (+-0.965%) : [≈]
                 crypto-md5.js | 21.582s ->  21.884s :  -1.399%  (+-0.493%) : [-]
                crypto-sha1.js |  9.764s ->   9.872s :  -1.106%  (+-0.367%) : [-]
          date-format-tofte.js |  2.198s ->   2.186s :  +0.546%  (+-1.142%) : [≈]
          date-format-xparb.js |  1.074s ->   1.066s :  +0.745%  (+-1.480%) : [≈]
                math-cordic.js |  3.098s ->   2.994s :  +3.357%  (+-0.464%) : [+]
          math-partial-sums.js |  1.698s ->   1.680s :  +1.060%  (+-0.537%) : [+]
         math-spectral-norm.js |  1.536s ->   1.506s :  +1.953%  (+-1.028%) : [+]
               string-fasta.js |  4.556s ->   4.462s :  +2.063%  (+-0.444%) : [+]
               Geometric mean: |                Speed up: 1.298% (+-0.252%) : [+]

[LaszloLango]

I think this should be return (value & (1ull << ECMA_VALUE_ERROR_POS)) != 0; like in other places in our code.

[LaszloLango]

Patch is updated and rebased to master.

[akosthekiss]

Why this warning? If value is not a boolean, does it cause any trouble (assertion, etc.)? I get from the code that the function will simply return false, wont it?

[LaszloLango]

To tell the truth I just copied&pasted the description from above (see: ecma_is_value_true). We can remove that warning message, it looks useless to me too.

[akosthekiss]

Made my observations above. (Those idiomatically used empty string error messages smell bad. However, it might need all-over-the-code changes to remove them properly, e.g., with functions that don't even expect a string, if they are not used whatsoever.) The rest is OK, as far as I can tell.

[LaszloLango]

@akiss77, maintaining and improving the error messages would be great and we need to do that. Do you mind if I raise an issue for that and do it in a separate PR?

[akosthekiss]

@LaszloLango, the patch is good with me as it is. And we don't necessarily have to come up with fancy error messages if we don't want to. But then we should declare that and then we can remove the string argument of those ecma_raise_* functions. The use of "" is what's bugging me.

[galpeter]

lgtm

[zherczeg]

I think we can improve error messages in a follow-up patch, this patch focuses on removing the completion value and already big enough.