extradoc / minute / pypy-sync-2005-10-20.txt

pypy-sync developer meeting 20th October

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


         Samuele Pedroni
         Armin Rigo
         Anders Chrigström
         Christian Tismer
         Holger Krekel 
         Eric van Riet Paap
         Michael Hudson
         Carl Friedrich Bolz
         Bert Freudenberg 
         Adrien di Mascio (late)

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 blockers

Topics of the week

Status update and next steps


Armin reports that classes/instance support in the rtyper for ootypes seems
reasonably complete now: class attributes, abstract base methods, function 
pointers stored on instances seem to work. Missing are all other high level
types like strings, lists, tuples, dicts, etc. Samuele notes that there might
still be problems with obscure prebuilt constants. Bert intends to work on
the squeak backend and Armin tries to help him as far as his work on reports
permit that. Holger notes that Impara's work should be more focused on
evaluating the possibilities how to integrate pypy and squeak to which Bert

stackless integration

Armin points at the discussion that took place on PyPy-dev:

and also notes that some more discussion might be needed. The current state of
the app-level stackless module is described at:

Christian is working on reducing the overhead of stackless. There was some 
discussion about the question whether stackless should be mentioned as
implemented in the current reports already. Holger and Carl Friedrich are
opposed to this idea since there will be a lot more reports coming.
Armin and Christian were in favor of writing about it. Some further discussion
is needed about this.


Carl Friedrich describes the current the of the l3interpreter: the
basic model is in place, for now only some operations with integers work.
There is some code that can convert low level to l3 graphs but it is far
from complete. Everybody agreed that nobody will work on the l3interpreter
until should not be worked on until the reports are finished.


Adrien thinks that compiler seems more or less stable now, although there are
still some bugs found every now and then. It was agreed upon that nobody will
work on it for a while.

javascript backend

Eric describes the current state of the JS backend: it passes around 50%
of the LLVM tests. The work on the JS backend is simpler because GC and
exception handling can be handed down to the underlying JS interpreter.
Eric will give a lightning talk on OSCON in Amsterdam and ask for feedback

Next pypy-sync meeting

Scheduled for next Thursday,  Oct. 27th 13.00h CET, conducted by Carl Friedrich


Carl Friedrich closes the meeting in time.

.. _`IRC-Log`:

Here is the full IRC log:: 

    **** BEGIN LOGGING AT Thu Oct 20 12:27:08 2005

    Oct 20 12:27:08 *	Now talking on #pypy-sync
    Oct 20 12:51:06 *	hpk ( has joined #pypy-sync
    Oct 20 12:51:17 *	pedronis ( has joined #pypy-sync
    Oct 20 12:51:57 *	stakkars ( has joined #pypy-sync
    Oct 20 12:58:57 *	mwh (n=mwh@ has joined #pypy-sync
    Oct 20 13:00:02 cfbolz	hi all! it 1pm on my clock now, shall we start?
    Oct 20 13:00:20 stakkars	DONE: more on source ordering, stackless experiments 
    Oct 20 13:00:21 stakkars	NEXT: finishing this, continue on spackless stuff
    Oct 20 13:00:21 stakkars	BLOCK: None
    Oct 20 13:00:43 mwh	last: form filling, recovering from sprint, playing with gcc optimisations
    Oct 20 13:00:43 mwh	next: reports, not sure what else
    Oct 20 13:00:43 mwh	blockers: german bureaucracy
    Oct 20 13:00:48 hpk	LAST: paris sprint, eu issues/background organisation, codespeak issues   NEXT: getting eric on board, some codespeak stuff, marocco   BLOCKERS: mixed connectivity
    Oct 20 13:01:04 *	arigo (n=arigo@ has joined #pypy-sync
    Oct 20 13:01:10 arigo	hi all
    Oct 20 13:01:10 cfbolz	LAST: sprinting, working on statistics gathering and code coverage
    Oct 20 13:01:10 cfbolz	NEXT: more statistics, code coverage, reports
    Oct 20 13:01:10 cfbolz	BLOCKERS: too many tasks :-)
    Oct 20 13:01:25 pedronis	LAST: sprint; looking into packing 0.6 test results, D04.4; annotation for fixed-size list support 
    Oct 20 13:01:27 pedronis	NEXT: reports, ...
    Oct 20 13:01:28 pedronis	BLOCKERS: -
    Oct 20 13:01:43 arigo	DONE: ootyper (classes/instances roughly complete now); some continuation
    Oct 20 13:01:43 arigo	experiments; more pypy-c debugging (it still doesn't run
    Oct 20 13:01:43 arigo	NEXT: reports D05.1 and D05.4 :-/
    Oct 20 13:01:43 arigo	BLOCKERS: -
    Oct 20 13:02:27 cfbolz	erics lines:
    Oct 20 13:02:31 cfbolz	Last: exception handling in Javascript
    Oct 20 13:02:31 cfbolz	Next: genjs
    Oct 20 13:02:31 cfbolz	Blockers: -
    Oct 20 13:02:44 cfbolz	arre?
    Oct 20 13:03:25 *	bertf ( has joined #pypy-sync
    Oct 20 13:03:35 *	ericvrp2 ( has joined #pypy-sync
    Oct 20 13:03:42 cfbolz	hi bert! want to post your acitvity report?
    Oct 20 13:03:48 cfbolz	hi eric! I just posted your lines
    Oct 20 13:03:49 stakkars	hi Bert, thanks for the wire
    Oct 20 13:03:56 ericvrp2	'kay
    Oct 20 13:04:17 bertf	activities - recovering from sprint
    Oct 20 13:05:11 cfbolz	there don't seem to be blockers that can be resolved
    Oct 20 13:05:20 cfbolz	let's proceed to the regular topics
    Oct 20 13:05:24 cfbolz	Status update and next steps 
    Oct 20 13:05:28 cfbolz	ootypessytem
    Oct 20 13:05:51 cfbolz	can somebody give an update about the status there?
    Oct 20 13:06:29 arigo	yes
    Oct 20 13:06:40 arigo	the classes/instances support seem reasonably complete now
    Oct 20 13:07:01 arigo	class attributes, abstract base methods, function pointers stored on instances, etc.
    Oct 20 13:07:12 arigo	I basically copied from lltype one test at a time
    Oct 20 13:07:29 bertf	nice :)
    Oct 20 13:07:38 arigo	so what's missing now are all other high-level types: strings, lists, tuples, dicts, etc.
    Oct 20 13:07:50 arigo	I guess we need to discuss more before we start that
    Oct 20 13:08:00 pedronis	and I suppose passing more obscure rpbc stuff
    Oct 20 13:08:07 cfbolz	arigo: indeed. later on #pypy?
    Oct 20 13:08:11 arigo	sure
    Oct 20 13:08:30 cfbolz	who is going to work on this the next weeks?
    Oct 20 13:08:31 arigo	pedronis: yes, though I think half of the obscure stuff will work magically :-)
    Oct 20 13:08:46 pedronis	yes, there's a chance
    Oct 20 13:08:48 arre	Sorry for joining late.
    Oct 20 13:08:53 *	stakkars is on a modem line. following on #pypy will be too expensive,today
    Oct 20 13:09:26 arigo	from now on, a lot of the hard stuff in will be easy and pushed to the back-end (i.e. Bert :-)
    Oct 20 13:09:39 bertf	fine with me ...
    Oct 20 13:09:43 cfbolz	ok
    Oct 20 13:10:03 arigo	I can continue to help anyway (reports reports reports)
    Oct 20 13:10:22 cfbolz	yes
    Oct 20 13:10:25 cfbolz	next topic?
    Oct 20 13:10:29 cfbolz	stackless integration
    Oct 20 13:10:29 cfbolz	---------------------
    Oct 20 13:10:33 cfbolz	status?
    Oct 20 13:10:40 arre	cfboltz: Want my lines?
    Oct 20 13:10:56 cfbolz	arre: ok
    Oct 20 13:11:08 arre	PREV: preparing data for reports, catching up on work done during sprint
    Oct 20 13:11:09 arre	NEXT: Help my parents move (No PyPy work)
    Oct 20 13:11:09 arre	BLOCKERS: None
    Oct 20 13:11:23 cfbolz	thanks
    Oct 20 13:11:25 arigo	stackless integration: more discussion needed with Christian
    Oct 20 13:11:45 *	hpk just sidenotes that the focus of impara should nevertheless probably be on an evaluation/laying out of the approaches of integrating squeak with pypy
    Oct 20 13:11:47 stakkars	well, the current plan looks fine.
    Oct 20 13:12:15 stakkars	I'm also looking into ways to reduce the overhead.
    Oct 20 13:12:30 *	bertf nods
    Oct 20 13:13:09 *	stakkars can discuss in the evening a bit
    Oct 20 13:13:19 cfbolz	current plan is the discussion that took on on pypy-dev?
    Oct 20 13:13:26 arigo	yes
    Oct 20 13:13:51 cfbolz	ok. christian, are you going to work on that?
    Oct 20 13:14:17 stakkars	sure. And I'd like to help with reporting (preferredly writing towads stackless stuff but well)
    Oct 20 13:14:30 cfbolz	great
    Oct 20 13:14:57 hpk	i think stackless is not the focus on phase 1 reports, but it can be hinted at when talking about flexibility 
    Oct 20 13:15:05 stakkars	after a break and some home trouble, I'm about to finish code ordering (convergence problem) and sqitch to stackless then
    Oct 20 13:15:12 *	adim ( has joined #pypy-sync
    Oct 20 13:15:13 arigo	hpk: D05.3 should talk about stackless in some details
    Oct 20 13:15:17 stakkars	ah - oh
    Oct 20 13:15:25 adim	Hi everyone (sorry for being late)
    Oct 20 13:15:36 hpk	(claiming that the architecture provides enoug leeway for stackless as if we hadn't tried it already, so we can be bold :-) 
    Oct 20 13:15:41 cfbolz	but this needs to be carefully done, since after all there are other stackless reports later
    Oct 20 13:16:10 cfbolz	so we probably don't want to give it all away already
    Oct 20 13:16:13 hpk	it should just be more or less bold claims / statements about the architecture for phase 1 IMO 
    Oct 20 13:16:19 stakkars	reporting about the current achievement would make sense,although I don't know whether Armin wants to do it
    Oct 20 13:16:51 arigo	maybe we can report about what we got at the end of the sprint
    Oct 20 13:16:53 hpk	no, i do think that we have to think about correct timing, our reports will always lack behind 
    Oct 20 13:17:11 stakkars	arigo: that's what I meant
    Oct 20 13:17:11 hpk	we have 40 reports to go (or some such) 
    Oct 20 13:17:22 arigo	stakkars: ok
    Oct 20 13:17:32 hpk	arigo, stakkars : but i disagree 
    Oct 20 13:17:36 arigo	hpk: I see the point
    Oct 20 13:18:11 cfbolz	let's defer this to #pypy, I'd say
    Oct 20 13:18:15 cfbolz	l3interpreter
    Oct 20 13:18:15 cfbolz	-------------
    Oct 20 13:18:28 cfbolz	I can say something about the status
    Oct 20 13:18:43 cfbolz	the basic model is in place, it ignores everything except integers for now
    Oct 20 13:18:43 hpk	cfbolz: defering without refering to persons seems difficult, but well 
    Oct 20 13:19:20 cfbolz	hpk: point taken 
    Oct 20 13:19:59 cfbolz	to l3interpreter there is some code that converts l2 to l3 but it can't do very much yet
    Oct 20 13:19:59 mwh	what types will the l3interpreter need to know about eventually?
    Oct 20 13:20:11 hpk	the question is (IMO) if we want to work much on l3interpreter in the next weeks 
    Oct 20 13:20:21 cfbolz	probably rather not
    Oct 20 13:20:56 cfbolz	too much work on reports to do
    Oct 20 13:21:11 mwh	indeed
    Oct 20 13:21:39 cfbolz	so I'd say I'll try to write a bit about the plans we had at the sprint and put it to sleep afterwards
    Oct 20 13:22:03 hpk	yip, i think so as well 
    Oct 20 13:22:15 cfbolz	ok then
    Oct 20 13:22:16 cfbolz	compiler
    Oct 20 13:22:16 cfbolz	--------
    Oct 20 13:22:24 cfbolz	could anybody tell what the status there is?
    Oct 20 13:22:38 mwh	well, we found another crazy bug last night
    Oct 20 13:22:53 adim	mwh: ah ?
    Oct 20 13:22:58 mwh	armin fixed it somewhat
    Oct 20 13:23:05 hpk	sorry, i have to ask: does everybody else think it's ok to postpone l3interpreter until  - say - december? 
    Oct 20 13:23:26 mwh	adim: from Lib/compiler, to do with __name mangling
    Oct 20 13:23:37 arigo	hpk: no, I'd only say until we are done with the bunch of thereports
    Oct 20 13:23:43 stakkars	yes. in effect this would continue at the next sprint, then.
    Oct 20 13:23:53 cfbolz	arigo: which will probably nearly be december :-)
    Oct 20 13:23:58 arigo	ok ok :-)
    Oct 20 13:24:02 hpk	arigo: ok, this will be aerly december, i am afraid, but we'll see 
    Oct 20 13:24:09 adim	cfbolz: for the compiler, I would say that we can consider it as finished since failing compliancy tests are note compiler-related bugs
    Oct 20 13:24:27 adim	but it doesn't guarantee that there's no bug left
    Oct 20 13:24:43 hpk	surely not
    Oct 20 13:24:55 cfbolz	so will anybody work on it?
    Oct 20 13:25:09 pedronis	arigo, hpk: I see, the important thing is that when we restart we have a short discussion on what next, if we manage before december
    Oct 20 13:25:10 arigo	the bugs are not known yet :-)
    Oct 20 13:25:13 adim	I don't really think so.
    Oct 20 13:25:16 mwh	as bugs are found, surely
    Oct 20 13:25:18 adim	:)
    Oct 20 13:25:28 cfbolz	ok, makes sense for everybody?
    Oct 20 13:25:39 hpk	makes sense to consider it finished for now, yes
    Oct 20 13:25:43 hpk	(to me) 
    Oct 20 13:25:55 cfbolz	to me as well
    Oct 20 13:26:15 arigo	yes
    Oct 20 13:26:21 stakkars	'k
    Oct 20 13:26:34 mwh	is this some kind of vote? :)
    Oct 20 13:26:37 cfbolz	yes
    Oct 20 13:26:39 cfbolz	:-)
    Oct 20 13:26:41 cfbolz	ok then
    Oct 20 13:26:42 *	mwh says "aye" too
    Oct 20 13:26:45 cfbolz	javascript backend 
    Oct 20 13:26:45 cfbolz	------------------
    Oct 20 13:26:50 arre	Aye.
    Oct 20 13:27:03 cfbolz	ericvrp2?
    Oct 20 13:27:03 ericvrp2	genjs is passing about 50% of the (llvm) unittests.
    Oct 20 13:27:04 ericvrp2	The resulting javascript code is run with a standlone js interpreter called Spidermonkey.
    Oct 20 13:27:06 hpk	mwh: we use to answer questions like this from the moderator :like "is everybody ..." :-) 
    Oct 20 13:27:06 ericvrp2	Actual testing in browsers is being worked on by Guido Wesdorp.
    Oct 20 13:27:09 ericvrp2	The code looks very similar but simpler that the C/LLVM code.
    Oct 20 13:27:12 ericvrp2	This is mostly because I can use JS' garbage collection and excption handling.
    Oct 20 13:27:17 ericvrp2	I am here now at euro OSCON in Amsterdam to give a 5 minute lightning talk about it
    Oct 20 13:27:18 ericvrp2	and my question to the audience will be to come up with use-cases.
    Oct 20 13:27:33 cfbolz	:-)
    Oct 20 13:27:38 hpk	:-)
    Oct 20 13:27:51 adim	:)
    Oct 20 13:27:53 *	ericvrp2 blurb
    Oct 20 13:28:05 hpk	does spidermonkey also provide some dom-simulation or so? 
    Oct 20 13:28:12 *	hpk notices that time is getting short
    Oct 20 13:28:20 ericvrp2	I don't think so
    Oct 20 13:28:22 cfbolz	indeed, less than two min left
    Oct 20 13:28:36 mwh	anything else?
    Oct 20 13:28:53 pedronis	should we think to do new a release?
    Oct 20 13:28:56 mwh	i probably won't sign my contract on this visit to duesseldorf :(
    Oct 20 13:29:00 cfbolz	:-(
    Oct 20 13:29:03 hpk	pedronis: next pypy-sync meeting? 
    Oct 20 13:29:07 cfbolz	indeed
    Oct 20 13:29:09 hpk	mwh: buah
    Oct 20 13:29:29 cfbolz	anybody opposed to close the meeting?
    Oct 20 13:29:36 hpk	nope
    Oct 20 13:29:58 stakkars	bye
    Oct 20 13:30:01 ericvrp2	bye
    Oct 20 13:30:04 adim	bye
    Oct 20 13:30:05 arigo	bye
    Oct 20 13:30:17 bertf	bye
    Oct 20 13:30:27 cfbolz	see you all next week!
    Oct 20 13:30:42 *	adim ( has left #pypy-sync
    Oct 20 13:30:57 *	mwh (n=mwh@ has left #pypy-sync
    Oct 20 13:31:09 *	ericvrp2 has quit ("leaving")
    Oct 20 13:31:12 *	arre ( has left #pypy-sync
    Oct 20 13:31:20 hpk	bye
    Oct 20 13:31:30 *	bertf ( has left #pypy-sync
    Oct 20 13:32:08 *	arigo (n=arigo@ has left #pypy-sync
    **** ENDING LOGGING AT Thu Oct 20 13:33:51 2005