[pypy-dev] Feature-comparatives benchmarks

Armin Rigo arigo at tunes.org
Sat Nov 5 10:16:54 CET 2005

Hi all,

Even though it is short, I think that the draft-low-level-encapsulation
would look complete enough if it had, for each described aspect:

* a link to the doc (I'll add them);

* some concrete benchmarks, in an "Experimental results" section,
  to see what would be the performance/memory impact of enabling each
  feature or leaving it out of the compiled PyPy.

As far as I know our benchmark results are not nicely regrouped at some
place, so I suggest we collect them on this mailing list.  We need:

* Stacklessness: what are the most up-to-date results?

* Multiple Interpreters: I guess no one tried so far to compile in a
  mode where the 'space' variables don't go away.  Let's.

* Concurrency: not much data here apart from the massive unexplained
  overhead of enabling threads.

* Memory Management: refcounting is veeery slow, which only shows that
  we could do much better there given sufficient efforts...

* Evaluation Strategy: the thunk object space.  Overhead?

A bientot,


More information about the Pypy-dev mailing list