extradoc / minute / pypy-sync-2005-09-15.txt

pypy-sync developer meeting 15th September 

Time & location: 1pm (30 minutes) at #pypy-sync 


         Samuele Pedroni
         Anders Lehmann
         Anders Chrigström
         Christian Tismer (minutes/moderation)
         Carl Friedrich Bolz
         Holger Krekel
         Eric van Riet Paap
         Niklaus Heidimann

Regular Topics 

- activity reports (3 prepared lines of info). 
  All Attendees submitted activity reports (see `IRC-Log`_ 
  at the end and 'LAST/NEXT/BLOCKERS' entries in particular)

- resolve conflicts/blockers
  No conflicts were discovered.

Topics of the week

A quick point about IRC logging

- There were no proposals for new channelsto be logged.
  The current set of logged channels is
  #pypy, #pypy-funding (password), #pypy-sync, #pypy-tb.

- There was no objection to have a second serverfor eggdrop.
  Holger added that he is going to work on some IRC infrastructure,
  without giving details. He proposed to use codespeak as
  the second server, which was agreed.

Participation on the CCC congress in Berlin

Holger would like to co-write a proposal and co-talk about PyPy.
Christian is considering to join this. We will ask Armin who was
interested but absent.

The suggested re-structuring of the selection process was not discussed.

Current optimization activities and allocation of work

There was consensus about current optimization efforts. There is no
obvious target visible that gives huge improvements. 
Opinions on profiling were different. We may need to have
profiling at various levels. But profiling might be unsharp,
if optimizing PyPy consists of a number of small improvements.
It was also proposed to judge optimization based upon the effect
multiplied by the speed of implementation.
It was concluded that we shouldfocus more on the reporting.
This topic will show up, again.

Other activities, follow-up on sync meeting of last week

Nobody showed up who wanted to do something last week, which resulted
in Armin, Samuele, Carl and Holger will work on that. It was proposed
to ask again and find a way to motivate each other to help with this.
Carl threw in that reporting actually can be fun. We should try to
find ways tomotivate each other on #pypy.

Making money out of PyPy

It was agreed that this topic is too big to be discussed on #pypy-sync.
Some interesting input about this can be found on the pypy list. See
"""Re: [pypy-dev] next pypy-sync meeting 20050915""" by Michal Wallace

Next pypy-sync meeting

Scheduled for next Thursday, Sept. 22nd, conducted by Anders Lehmann.
The timing issue was touched, again, but needs to be deferred.


Christian closes the meeting in time at 13:33pm.

The IRC log can be found at

Holger asked to include it in the log, anyway, for ease of reference.

.. _`IRC-log`: 

Here is the full IRC log. Note that time is given in UTC:: 

  [11:31] aleale ( joined #pypy-sync.
  [11:48] pedronis ( joined #pypy-sync.
  [11:51] stakkars ( joined #pypy-sync.
  [11:52] arre ( joined #pypy-sync.
  [11:55] hpk ( joined #pypy-sync.
  [11:55] ericvrp ( joined #pypy-sync.
  [11:55] cfbolz ( joined #pypy-sync.
  [12:02] <hpk> hello all
  [12:02] Action: hpk notices it's 13:03 on his clock
  [12:02] <stakkars> hi all! Yes, I'm somewhat waiting for Armin and Michael
  [12:03] nik ( joined #pypy-sync.
  [12:03] <stakkars> I'd say let's start with the regular part
  [12:03] <stakkars> ok?
  [12:03] <hpk> armin may not come (and we should not wait too much in general i think)
  [12:03] <hpk> yes
  [12:03] <cfbolz> fine with me
  [12:03] <ericvrp> great
  [12:03] <ericvrp> hi all
  [12:03] <stakkars> then I start with my three lines:
  [12:04] <stakkars> DONE: some "ten lines for ten percent" speedups on dicts, interning would take more to do right
  [12:04] <stakkars> NEXT: working on different approaches to reduces PyPy's stack requirements
  [12:04] <stakkars> BLOCK: demotivation by randomness of optimization criteria
  [12:04] <cfbolz> DONE: non-pypy
  [12:04] <cfbolz> NEXT: non-pypy, vacation
  [12:04] <cfbolz> BLOCKERS: none
  [12:04] <aleale> prev: translate_pypy_new, internal DFKI stuff
  [12:04] <aleale> next: see my family, moderate #pypy-sync
  [12:04] <aleale> blockers: -
  [12:04] <arre> PREV: astcompiler
  [12:04] <arre> NEXT: astcompiler
  [12:04] <arre> BLOCKERS: None
  [12:04] <ericvrp> last: refactoring
  [12:04] <ericvrp> next: profiling and dive into transformations
  [12:04] <ericvrp> blockers: -
  [12:05] <hpk> last: EU issues, background communication
  [12:05] <hpk> next: non-pypy issues, EU-issues
  [12:05] <hpk> blockers: -
  [12:05] <pedronis> Last: astcompiler, rtyper ann fixes, hlinvoke support to move dict impl to low-level,
  [12:05] <pedronis> some writing, understanding the fine point of the flow space again
  [12:05] <pedronis> Next: astcompiler, reports, translate_pypy, ?
  [12:05] <pedronis> Blockers: -
  [12:05] <nik> last: non-pypy
  [12:05] <nik> next: non-pypy
  [12:05] <nik> blockers: -
  [12:05] <hpk> nik: blockers: non-pypy? :-)
  [12:05] <nik> uhm, yep ;)
  [12:06] <stakkars> ok, then I'd add my toothache tomy blockers :-)
  [12:06] <stakkars> anything else?
  [12:07] <stakkars> I don't see blockers which are not touched by other topics.
  [12:07] <stakkars> topic 1:
  [12:07] <stakkars> A quick point about IRC logging
  [12:07] <stakkars> does anybody have proposals about that to log else?
  [12:07] <hpk> let me note that i plan to setup some IRC-logging infrastructure
  [12:07] <hpk> on codespeak at some point.
  [12:07] <stakkars> do we want a second server in another network?
  [12:07] <stakkars> ah
  [12:07] <hpk> but that shouldn't interfere with your nice efforts. 
  [12:08] <hpk> it probably should be the second server then. 
  [12:08] <stakkars> codespeak is of course a possible target. You can then easily attach something to thebasic logs
  [12:08] <hpk> the HHU server is probably more helpful for getting computing/testing power
  [12:08] <stakkars> ok, then I can do an eggdrop setup. Some postprocessing of the logs would be really really good!
  [12:09] <stakkars> the performance penalty of eggdrop is very low.
  [12:09] <stakkars> the criteria should be reliability of network and server.
  [12:09] <hpk> yip.
  [12:10] <stakkars> I take it that we want codespeakto be a second server, and there are no proposals of other channels?
  [12:10] <stakkars> anything to be said, or next topic?
  [12:10] <cfbolz> next for me
  [12:10] <hpk> yip. next for me as well.
  [12:10] <aleale> next
  [12:10] <arre> next.
  [12:11] <stakkars> Participation on the CCC congress in Berlin
  [12:11] <hpk> i'd like to co-write a proposal and co-talk about PyPy.
  [12:11] <stakkars> besides discussion of who wants to show what, I'd like to repeat: K can helptoget you a room
  [12:11] <cfbolz> I am interested in that
  [12:11] <hpk> great, but i know very many people in berlin :_) 
  [12:12] <hpk> basically the CCC congress is much about showing off/reporting about cool technical stuff
  [12:12] <stakkars> I was offering help, and you are overloaded, so let's put it together
  [12:13] <hpk> not so much about scientific musings about possible benefits :-)
  [12:13] <stakkars> did everybody read Armin's proposal?
  [12:13] <nik> which proposal?
  [12:14] <stakkars> draft-file in doc
  [12:14] <hpk> you mean the report/deliverable draft? 
  [12:14] <stakkars> don't read it now,toolong
  [12:14] <hpk> is that much related to the CCC congress? 
  [12:14] <stakkars> draft-dynamic-language-translation.txt
  [12:14] <pedronis> is not a talk proposal
  [12:14] <stakkars> I think this wasmeant as a start for a talk on CCC?
  [12:15] <hpk> no, for a start ont he EU reporting issues
  [12:15] <pedronis> it's the start of one of our reports
  [12:15] <stakkars> uppsi :-)
  [12:15] <pedronis> I even wrote stuff in it
  [12:15] <hpk> np
  [12:15] <stakkars> pedronis: I know
  [12:16] <hpk> stakkars: you want to go and talk at CCC? 
  [12:16] <stakkars> I'm not really sure, but I thought of it,not only becaue I was very quiet, yet.
  [12:17] <stakkars> no idea if they want more than one talk. I might offer to do something with you if you like.
  [12:17] mwh ( joined #pypy-sync.
  [12:17] <hpk> stakkars: yes, that's fine 
  [12:17] <hpk> stakkars: also we should ask armin who was interested 
  [12:17] <hpk> i don't think it's a problem to do a one hour talk with three persons
  [12:17] <hpk> we have done that before
  [12:17] <stakkars> let's try if we are able to share:
  [12:17] <hpk> we can also consider doing two talks (but now now i suggest)
  [12:18] <hpk> s/now now/not now/
  [12:18] <stakkars> I think we should move on with the topics.
  [12:18] <hpk> yip from my side.
  [12:18] <stakkars> ok?
  [12:18] <stakkars> Current optimization activities and allocation of work
  [12:18] <cfbolz> yes
  [12:18] <stakkars> we have done some random improvements, where none of these but the call optimization
  [12:19] <stakkars> came in any waynear the criteria of a2.0 speedup.
  [12:19] <stakkars> We might think to drop optimizationj completely at this point, or change criteria.
  [12:19] <hpk> indeed. i think - as mwh points out on pypy-dev - we should go for profiling rather than direct optimizations at the moment
  [12:20] <hpk> having profiling at various levels provides better grounds for judging optimizations later on 
  [12:20] <pedronis> well, we should work on the reports more too
  [12:20] <stakkars> I also would like to add three lines:
  [12:20] <stakkars> we are very unlikely to hit such huge speed gains again at all.
  [12:20] <hpk> pedronis: yes. but that's a different topic :-) 
  [12:20] <stakkars> ANd if we have 10 percent speedups, it takes just 40 of these to rech the goal.
  [12:21] <stakkars> >>> 1.1 ** 40
  [12:21] <stakkars> 45.259255568176101
  [12:21] <cfbolz> :-)
  [12:21] <cfbolz> if we do one a week...
  [12:21] <cfbolz> (just kidding, of course)
  [12:22] <hpk> well, what about going for profiling rather than optimizations? 
  [12:22] <stakkars> other thing: multiply time effort to do the implementation by it's effect, that would be a good criteria.
  [12:22] <stakkars> hpk: I fear that profiling will not be very sharp, as it was not yet.
  [12:22] <stakkars> The problem is that we loose speedd in many tiny issues, and we needto go the hard wayto remove one by one.
  [12:22] <arre> I'm all for profiling of various sorts.
  [12:22] <stakkars> absolutely, anecessary companion.
  [12:23] <stakkars> may I conclude that we need to concentrate on reporting ATM and don't waste time on optimization which needs much resources?
  [12:24] <stakkars> anything on this, or next topic?  ( 5 min left)
  [12:24] <hpk> at this point, we cannot all focus just on one thing
  [12:24] <hpk> reporting is currently thought to be brought forward by armin, samuele, carl and me 
  [12:24] <hpk> i do think that help is much appreciated 
  [12:24] <stakkars> that's next topic
  [12:24] <stakkars> Other activities, follow-up on sync meeting of last week
  [12:25] <stakkars> do we realy stick with last week's conclusion, or should we contractmore people on this
  [12:25] <cfbolz> nobody said he wanted to something last week
  [12:26] <stakkars> yes, I think that's bad. We need to structure work in a way that it is ieasy to work on something, and it is no func
  [12:26] <stakkars> not much fun, so we need to motivate each other.
  [12:26] <hpk> :-)
  [12:27] <cfbolz> I actually think reporting _can_ be fun :-)
  [12:27] <stakkars> I think this is great news, please tell this on #pypy, we are almost finished here
  [12:27] <stakkars> Making money out of PyPy
  [12:28] <stakkars> something interesting has been said on the list, and the topic is too big for pypy-sync, right?
  [12:28] <pedronis> yes is too big
  [12:28] <aleale> yes
  [12:28] <hpk> i guess, it's a somewhat timeless question probably better discussed at some pub
  [12:28] <nik> #pypy-pub? ;)
  [12:29] <stakkars> my opinion was to get it a bit more pressure
  [12:29] <hpk> :-)
  [12:29] <stakkars> good idea.
  [12:29] <stakkars> we have 3 mins left, becasue we started 3 after 1pm
  [12:29] <stakkars> or we can close now if you like
  [12:30] <hpk> fine for me.
  [12:31] <ericvrp> ok then, till next weak. same place, same time?
  [12:31] <ericvrp> s/weak/week/ ehum :)
  [12:31] <stakkars> then thank you all for the nice input, closing JIT
  [12:31] <hpk> yip, actually that was a topic we had from last time, the timing issue
  [12:31] <hpk> stakkars: great, thanks
  [12:31] <cfbolz> see you then
  [12:31] cfbolz ( left #pypy-sync ("Verlassend").
  [12:31] <ericvrp> bye
  [12:31] <aleale> see you next week
  [12:32] ericvrp ( left #pypy-sync.
  [12:32] <stakkars> note that the protocol will not include the IRC log.
  [12:32] <nik> see you
  [12:32] nik ( left #pypy-sync.
  [12:32] <stakkars> it can be found at
  [12:32] stakkars ( left #pypy-sync.
  [12:32] arre ( left #pypy-sync.
  [12:32] aleale ( left #pypy-sync.