Java > Scala

[Somedude]

Le 03/12/2011 10:02, Russel Winder a écrit :
> On Fri, 2011-12-02 at 11:51 -0800, Walter Bright wrote:
> [...]
>> The charts on that page refuse to display in IE.
> 
> You probably need to tell them about this, they don't have resources to
> check stuff other than cursorily, they rely on user feedback.
> 
>> Nevertheless, it doesn't seem to compare against C, but against CPython.
> 
> I think I may be responsible for a slight
> misunderstanding/misrepresentation here. Apologies.
> 
> PyPys main short term goal is to become the best realization of the
> Python standard; to supplant CPython as the reference implementation.  A
> few weeks ago Laura Creighton announced on the PyPy Dev email list that
> they had achieved the "5 times faster than CPython" goal on the critical
> benchmarks.  The comparison is the RPython implementation of Python vs
> the CPython implementation.
> 
> It is an avowed goal of the PyPy project that its performance should be
> close enough to C that people use PyPy rather than C.  Currently people
> can use Cython to write (slightly annotated) Python code that gets
> compiled to C and thence to native code to get performance as close to C
> as makes no difference.  The intention is to avoid having to annotate
> and compile to get the performance.  This is a long term goal for PyPy
> and not yet near fruition, hence people using Cythoin for the
> performance critical sections.
> 
> This is an area where Go and PyPy will be fighting head on.
> 
>>> Not exactly rigorous mathematical proof, but pretty strong evidence. Also, I use
>>> PyPy once in a while for projects where speed matters a little but I want to
>>> share my code with Python people or want to use Python's huge standard library.
>>> Anecdotally, it's definitely faster. The reason has nothing to do with the
>>> language it's written in. It's because PyPy JIT compiles a lot of the Python
>>> code instead of interpreting it.
> 
> The PyPy JIT is clearly a "big win".  I am sure Armin will come up with
> more stuff :-)
> 

Well, right now, Pypy is indeed 5 times faster than CPython, but it's
still about 4 times slower than LuaJit and Javascript v8, which are
comparable in speed with Java (especially LuaJit). And basically, this
sort of speed is the best that has been done with JIT so far and it
seems to be quite hard to improve on.

So maybe they can hope to reach LuaJIT speed in the near future, but
before reaching C speed, I suppose there is a long way to go.