pypy-sync developer meeting 1st September
Time & location: 1pm (30 minutes) at #pypy-sync
Adrien Di Mascio,
Armin Rigo, Carl Friedrich Bolz (later)
Holger Krekel (minutes/moderation)
Eric van Riet Paap
- 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
Recap of discussed development areas until October
Everyone agreed that we want to deal with cleanups
and simplification refactorings until the Paris
Sprint rather than optimizations. Optimizations should only
be tackled for "low hanging fruits" which is defined to have
a good ratio of LOCs against speedup. For example,
the recent specializations of function calls which lead
to a speed-up by 2-3 times were considered as having
a good ratio. Moreover we looked in more detail
at the following cleanup/refactoring issues:
- bringing the compiler to good state. Ludovic and
Adrien with help from Samuele and Arre will work
on making the current 'astcompiler' the (translateable)
and compliant default compiler.
- the translation driving (translate_pypy.py and friends)
Anders Lehmann is going to work on this and Armin has
some refactored code and intends to help Anders along
with Samuele. Eric will keep an eye on having LLVM
- improved test/reporting support. Holger intends to
work on this.
Paris Sprint announcement
Ludovic is going to prepare a Paris Sprint announcement
tomorrow with help from Holger. The meeting reconfirms
the following sprint topics:
- threading and GC
- refactoring/translation features
- start on JIT/optimizations/stackless
Preparation/Moderation Next pypy-sync meeting
Armin, Anders L. and Christian are ready to prepare
the next three pypy-sync meetings, respectively:
8th September: Armin
15th September: Christian
22rd September: Anders L.
We'll see who is going to do the next pypy-sync
meetings after that.
Holger closes the meeting in time at 13:26pm.
Here is the full IRC log::
**** BEGIN LOGGING AT Thu Aug 18 12:59:41 2005
Sep 01 12:53:14 --> You are now talking on #pypy-sync
Sep 01 12:56:18 --> pedronis (n=Samuele_@c-398b70d5.022-54-67626719.cust.bredbandsbolaget.se) has joined #pypy-sync
Sep 01 12:56:24 --> stakkars (i=pyanan@i577B4CC6.versanet.de) has joined #pypy-sync
Sep 01 13:02:03 <hpk> i guess we start (hope arigo drops over soon)
Sep 01 13:02:18 <hpk> here is the agenda (as mailed yesterday):
Sep 01 13:02:21 <hpk> - activity reports (3 prepared lines of info).
Sep 01 13:02:21 <hpk> - resolve conflicts/blockers
Sep 01 13:02:21 <hpk> *Topics of the week*
Sep 01 13:02:21 <hpk> - Recap of discussed development areas until October
Sep 01 13:02:21 <hpk> - Paris Sprint announcement
Sep 01 13:02:21 <hpk> - Preparation/Moderation next pypy-sync meeting
Sep 01 13:02:22 --> ludal (firstname.lastname@example.org) has joined #pypy-sync
Sep 01 13:02:53 <hpk> let's start with activity reports in this order:
Sep 01 13:03:02 <hpk> adim,aleale,ericvrp,hpk,ludal,pedronis,stakkars
Sep 01 13:03:24 <adim> LAST: none
Sep 01 13:03:24 <adim> NEXT: astcompiler
Sep 01 13:03:24 <adim> BLOCKERS: none
Sep 01 13:03:28 <aleale> This week: recover from Heidelberg, clean-up _codecs, planning of wp 9 and 10
Sep 01 13:03:30 <stakkars> how about reverse,mine is in the clipboad
Sep 01 13:03:37 <aleale> Next week: more cleanups, more planning, some compliancy issues
Sep 01 13:03:45 <aleale> Blockers : -
Sep 01 13:03:52 <hpk> LAST: heidelberg sprint, reporting, mailing
Sep 01 13:03:52 <hpk> NEXT: steps towards better test reporting and general refactoring
Sep 01 13:03:52 <hpk> BLOCKERS: None
Sep 01 13:04:03 <ericvrp> last: slotified llvm backend, bugfixes, added gc atomic calls to genc
Sep 01 13:04:04 <ericvrp> next: test experimental llvm escape analysis
Sep 01 13:04:06 <ericvrp> blockers: -
Sep 01 13:04:30 <hpk> stakkars: go ahead if you can't keep your clipboard content :)
Sep 01 13:04:30 <pedronis> LAST: sprint, small fixes, looked a bit at performance
Sep 01 13:04:32 <pedronis> NEXT: report planning with Armin, cleanups/help with compiler
Sep 01 13:04:33 <pedronis> BLOCKERS: how we want to distribute cleanups work/design
Sep 01 13:04:55 <ludal> Last week: sprint then none
Sep 01 13:04:55 <ludal> Next week: astcompiler
Sep 01 13:04:55 <ludal> blockers: none
Sep 01 13:05:18 <stakkars> DONE: reimplemented list overallocation in low-level, removing it from listobject
Sep 01 13:05:25 <stakkars> NEXT: making dictionaries low-level, maybe looking into trusted refcounting.
Sep 01 13:05:31 <stakkars> BLOCK: maybe a missing interface between lowlevel types that prevents malicious use of fields
Sep 01 13:06:07 <hpk> ok, pedronis's blocker is a topic of this meeting
Sep 01 13:06:19 <hpk> and stakkars blocker should be discussed on #pypy i guess
Sep 01 13:06:40 <stakkars> yes
Sep 01 13:06:49 <hpk> ok, first topic of the week:
Sep 01 13:06:54 <hpk> Recap of discussed development areas until October
Sep 01 13:07:14 <hpk> we said that we want to focus on cleanups/refactoring until we start with otpm,ization wp06-XX efforts in october
Sep 01 13:07:29 <hpk> the three areas i wrote down at last sprint:
Sep 01 13:07:33 <hpk> compiler
Sep 01 13:07:34 <hpk> translation driving
Sep 01 13:07:42 <hpk> testing stuff
Sep 01 13:07:51 <hpk> is there anything that is missing or different opinions?
Sep 01 13:08:13 <aleale> consolodating the use of _codecs
Sep 01 13:08:28 <hpk> in which sense?
Sep 01 13:08:41 <aleale> The same algoritm of escaped strings is at least 33 places
Sep 01 13:08:49 <aleale> s/33/3
Sep 01 13:09:18 <aleale> there might be more
Sep 01 13:09:20 <pedronis> at least the string one needs to be at interp-level for bootstrapping reason
Sep 01 13:09:47 <aleale> ok lets defer it #pypy
Sep 01 13:09:52 <hpk> ok
Sep 01 13:09:53 --> arigo (n=arigo@pdpc/supporter/sustaining/arigo) has joined #pypy-sync
Sep 01 13:10:12 <hpk> ludal,adim: you are aware that arre and samuele are willing to help with astcompiler and the compiler efforts?
Sep 01 13:10:31 <ludal> now yes :)
Sep 01 13:10:49 <hpk> ludal: it also is in the technical board minutes from heidelberg
Sep 01 13:11:00 <hpk> jsut so you know
Sep 01 13:11:14 <ludal> it needs to become translatable, and probably we'll then have to repass the compliance tests
Sep 01 13:11:28 <hpk> stakkars: so simplifying/unifying our dict implementations also goes in the camp of refactoring/cleanups i guess
Sep 01 13:11:45 <stakkars> quick question: how clear is what we are doing when? Do we want to work sequentially or do some
Sep 01 13:11:58 <stakkars> optimization in parallel? I'm asking since this has happened already.
Sep 01 13:12:15 <hpk> i know, i am more in the camp of not tackling too much optimization
Sep 01 13:12:32 <hpk> but rather get our code base into better shape
Sep 01 13:13:04 <stakkars> I think there are different opinions form what I see happending. We should be clear about this.
Sep 01 13:13:07 <hpk> it is open who is going to work on improving translate_pypy and friends
Sep 01 13:13:22 <pedronis> the optimisation that was done had an enormous number of lines/efforts speed up ratio
Sep 01 13:13:37 <pedronis> unless there is some other such very low hanging fruit
Sep 01 13:13:50 <pedronis> I think cleanups take precedence
Sep 01 13:13:57 <hpk> yes, i agree
Sep 01 13:14:12 <pedronis> othewise they risk to never to be done
Sep 01 13:14:12 <stakkars> I se, it was the "very much issue". Then I will trash my list overallocation.
Sep 01 13:14:28 <hpk> everyone else agrees with "only very good ratios of LoC/speed ups" are ok?
Sep 01 13:14:32 <ludal> maybe looking at the possible optimization would help coming up with a more flexible design
Sep 01 13:14:51 <pedronis> stakkars: ?
Sep 01 13:15:02 <pedronis> trashing done work doesn't seem a good idea
Sep 01 13:15:12 <stakkars> at least we should do considerations. I found some corner cases on list allocation,
Sep 01 13:15:23 <aleale> I aggree to only addres low hanging fruits at this time
Sep 01 13:15:30 <stakkars> which need a bit of discussion, because optimization is hard wihtout knowledge of GC
Sep 01 13:16:01 <hpk> is anyone here ready to tackle translate_pypy cleanups?
Sep 01 13:16:02 <stakkars> low-hanging fruit is ok. What do I do if the fruit wasn't that sleshy, after all? :-)
Sep 01 13:16:08 <stakkars> yes
Sep 01 13:16:36 <aleale> hpk: yes
Sep 01 13:16:47 <arigo> I started some time ago something about cleaning up Translator
Sep 01 13:17:00 <arigo> but I'm not sure I'd like to finish it right now (it looks like work)
Sep 01 13:17:15 <hpk> arigo: :-)
Sep 01 13:17:22 <arigo> for me, outlining the reports' content and starting on them is higher priority
Sep 01 13:17:38 <stakkars> on cleaning up dictionaries, this gives both speed and nicer source.Postpone as well?
Sep 01 13:17:39 <aleale> makes sense
Sep 01 13:17:55 <hpk> right, ok. then aleale can give it a go and whoever wants to join.
Sep 01 13:18:12 <pedronis> I can help
Sep 01 13:18:17 <hpk> great
Sep 01 13:18:20 <hpk> i think we can close this topic now (there also is and should be discussion on pypy-dev regarding some of the topics)
Sep 01 13:18:26 <pedronis> but it has less priority than reports and compiler
Sep 01 13:18:27 <ericvrp> I will keep checking if it's still llvm compatible
Sep 01 13:18:34 <hpk> pedronis: sure
Sep 01 13:18:35 <stakkars> ok, I misunderstood the sudden massive speedup-checkins, it was just a big fruit, right?
Sep 01 13:18:51 <hpk> yes, a factor of 2-3 speedup in pystones
Sep 01 13:19:03 <pedronis> and richards
Sep 01 13:19:06 <stakkars> (and I got just 5 percent)
Sep 01 13:19:11 <arigo> pedronis: what do you mean by compiler ?
Sep 01 13:19:27 <hpk> arigo: we discussed this earlier, the translateable astcompiler
Sep 01 13:19:28 <pedronis> helping logilab making it translatable
Sep 01 13:19:30 <ludal> translate it
Sep 01 13:19:34 <arigo> hpk: ah, sorry of course
Sep 01 13:19:40 <hpk> please next topic: Paris Sprint announcement
Sep 01 13:20:04 <hpk> adim, ludal: do you intend to draft up an annoucnement in extradoc/sprintinfo/paris-annoucnement.txt?
Sep 01 13:20:20 <ludal> yes, I'll do that tomorrow
Sep 01 13:20:35 <hpk> ok, i can help you a bit.
Sep 01 13:20:44 <hpk> do we still agree on the topics?
Sep 01 13:20:50 <hpk> - threading and GC
Sep 01 13:20:50 <hpk> - refactoring/translation features
Sep 01 13:20:50 <hpk> - start discussing/heading for JIT/optimizations/stackless
Sep 01 13:20:53 <ludal> thanks
Sep 01 13:21:26 <arigo> yes
Sep 01 13:21:31 <aleale> yes
Sep 01 13:21:40 <adim> seems good to me
Sep 01 13:21:54 <ludal> yes
Sep 01 13:22:16 <arigo> "finish up reports in a hurry" shouldn't probably be announced :-)
Sep 01 13:22:24 <pedronis> :)
Sep 01 13:22:37 <hpk> one more thing i'd like to consider (but it needn't go into the announcement) is the issue of easing porting of C-modules for developers new to pypy
Sep 01 13:22:54 <pedronis> yes, also refactoring Translator is probably related what we want to do in phase2
Sep 01 13:23:01 <pedronis> so maybe it can partly wait for paris
Sep 01 13:23:01 <hpk> yes
Sep 01 13:23:28 <hpk> ok, that's it for the paris sprint/annoucnement for now, i guess.
Sep 01 13:23:42 <hpk> last topic: Preparation/Moderation Next pypy-sync meeting
Sep 01 13:24:11 --> cfbolz (email@example.comWays.net) has joined #pypy-sync
Sep 01 13:24:14 <hpk> who would like to do it the next time? (Can be multiple people so we can rotate a bit)
Sep 01 13:24:18 <hpk> cfbolz: good morning
Sep 01 13:24:26 <arigo> I could do it
Sep 01 13:24:28 <cfbolz> sorry
Sep 01 13:24:35 <stakkars> i can do it.
Sep 01 13:24:42 <aleale> I'd like to volonteer
Sep 01 13:25:03 <stakkars> ok, let's pick a random number.
Sep 01 13:25:08 <ericvrp> 42
Sep 01 13:25:17 <stakkars> 49. mine is higher.
Sep 01 13:25:29 <hpk> great, then maybe just arigo, stakkars, aleale in that order?
Sep 01 13:25:40 <hpk> let me just note that so far i had to come up with topics for the meetings myself
Sep 01 13:25:46 <stakkars> wise man spoke :-)
Sep 01 13:26:39 <stakkars> 4 minutes left, is there some opne point?
Sep 01 13:26:42 <arigo> ok, then I will do the next week one
Sep 01 13:27:00 <hpk> and it would probably help the new moderators if at least sometimes topics are pushed to them
Sep 01 13:27:10 <arigo> indeed :-)
Sep 01 13:27:28 <hpk> but part of the game is identifying what would benefit from pypy-sync attention and decisions
Sep 01 13:27:37 <hpk> :-)
Sep 01 13:27:52 <hpk> ok, then let me close my last pypy-sync meeting for a few weeks :-)
Sep 01 13:28:06 * arigo claps
Sep 01 13:28:08 <cfbolz> should I post my lines?
Sep 01 13:28:17 <stakkars> do it,not ask
Sep 01 13:28:19 <cfbolz> LAST: worked on/finished summer of code project on GC
Sep 01 13:28:19 <cfbolz> NEXT: PyPy unrelated stuff (another exam)
Sep 01 13:28:19 <cfbolz> BLOCKER: some sort of strange behaviour under python 2.3
Sep 01 13:28:33 <hpk> cfbolz: congrats, btw!
Sep 01 13:28:59 <stakkars> cfbolz: please tell about the behavior on #pypy
Sep 01 13:29:04 <cfbolz> :-)
Sep 01 13:29:35 <cfbolz> ok
Sep 01 13:29:39 <-- cfbolz (firstname.lastname@example.orgWays.net) has left #pypy-sync ("Verlassend")