pypy-sync developer meeting 2nd February 2006
Time & location: 1pm (30 minutes), GMT+1 at #pypy-sync
Adrien di Mascio
Eric van Riet Paap
Carl Friedrich Bolz
Armin Rigo (moderation & minutes)
- activity reports (3 prepared lines of info).
All Attendees submitted activity reports (see `IRC-Log`_
and 'LAST/NEXT/BLOCKERS' entries in particular)
- resolve conflicts/blockers: No blockers
Topics of the week
PyCon sprint announcement
Discussed and finalized the topics (in no particular order):
* py lib subtrack, with a focus on issues relevant to PyPy
* write GCs (we might start translating our own GCs by then)
* rctypes (newcomer-friendly: writing 'socket' or some other module in ctypes)
* logic programming: constraints; adding dataflow variables to PyPy
* general JIT stuff (maybe)
* general experiments with app-level code using PyPy features (thunk, ...)
What could be better in py.test from the point of view of testing PyPy?
* produces far too much output in general
* improve the test selection mecanisms, e.g. -k should allow us to select
based on class name, or select one test of a generative test, etc.
* test coverage
* doctests (important from a community point of view)
Complete IRC log
<adim> Hi everyone
<cfbolz> hi all!
--> mwh (email@example.com) has joined #pypy-sync
<arigo> hi all
<arigo> let's start, please post your three-liners...
<ericvrp> LAST: raisingop2direct_call transformation, NEXT: finish transformation, BLOCKERS: -
<hpk> LAST: codespeak migration, wp14, non-pypy NEXT: Ireland workshops BLOCKERS: no
<auc> last : almost "finished" our prototype oz-like computation space
<auc> next : add merge op, implement basic search strategies with it
<auc> block : nil
<adim> LAST: helped Nicolas to prepare the "SolutionsLinux" exhibition / talked about PyPy to everyone passing around the Logilab's stand
<adim> NEXT: restart aspect / WP10 stuff
<adim> BLOCKERS: none
<mwh> LAST: sprint, being ill,
<aleale> PREV: Sprint on Logic, trying pyontology on an inhouse ontology
<aleale> NEXT: Pyontology, documentation, a little logic
<aleale> BLOCKERS : -
<cfbolz> LAST: mallorca sprint, some small pickling experiments
<cfbolz> NEXT: finish mailwitness sprint, relax a bit, continue gc work, py-lib release
<cfbolz> BLOCKERS: too much to do, as usual
<mwh> NEXT: finish genc refactoring
<mwh> BLOCKERS: -
<arre> PREV: Recuperating after Mallorca sprint.
<arre> NEXT: Working on the JIT with Samuele.
<arre> BLOCKERS: None
<pedronis> LAST: sprint, hint annotation
<pedronis> NEXT: more jit
<pedronis> BLOCKERS: -
<arigo> LAST: sprint, no much; NEXT: some JIT work; BLOCKERS: still recovering from sprint
<arigo> PyCon sprint announcement
<arigo> during the Mallorca sprint, we wrote a file "what to do next"
<arigo> this contains several topics that make sense to include in the PyCon sprint announcement
<arigo> I think that these topics could all be mentioned in the announcement,
<arigo> but it would be nice to have a few more
<hpk> we should hint at the pypy talks during the pycon talks in the announcement
<hpk> and also (i mentioned this to arigo already) i'd like to do a bit of py lib sprinting as a sub-track
<hpk> related to improvements visible for pypy development
<mwh> we should try hard to make the sprint announcements non-intimidating
<arigo> mwh: yes
* hpk thinks that the py lib topics can help there :)
<arigo> I expect the sprint to be pretty much a "what are you interested in?" kind of sprint, with not too many pypy corers there
<cfbolz> how many days are there actually?
<cfbolz> so it's not _that_ big anywya
<mwh> but for one reason and another a lot of people will only be there for 3.5 or so days
<hpk> the sprint directly starts after the conf, btw (no break day)
<pedronis> which means tired people
<hpk> a bit, yes
<hpk> does it make sense to offer experimentation stuff?
<hpk> like extending the thunk space, playing on top of it, refining it?
* hpk thinks that we will need an internal core sprint between now and may because both japan and pycon are likely not allowing us to tackle the tough stuff
<arigo> we could propose this kind of things
<mwh> i think the gc stuff should be mentioned
<arigo> also with other spaces, not just thunk
<mwh> particularly if you can translate the gc framework by then...
<hpk> mwh: definitely
<cfbolz> mwh: which is unclear, but indeed the goal :-)
<hpk> being able to write GCs would be cool - but i am a bit skeptical (without really being able to judge)
<mwh> cfbolz: :)
<arigo> hpk: I'm more optimistic
<hpk> good :)
<arigo> anyway, it's a good topic, yes
<mwh> hpk: can an additional sprint be another topic ?
<arigo> and definitely rctypes
<hpk> are the rctypes people from mallorca going to be at pycon?
<hpk> i mean gromit and stephan?
<arigo> I guess not
<arigo> that shouldn't stop us, though
<cfbolz> mwh: indeed :-)
<hpk> arigo: ok
<arigo> writing an ext module in ctypes is a good way to contribute to PyPy
<mwh> arigo: yes
<arigo> because learning internals for 4 days isn't quite enough
<aleale> I hope to come so we could add a topic about implementing logic
<arigo> aleale: anything more precise in mind?
<hpk> is socket still a topic or only after we know that we will build on the (r)ctypes approach?
<arigo> I believe we can make socket-on-ctypes a topic, but it also seems to be a very demanded topic
<aleale> well it is hard to say since I dontknow how far we will be when Pycon comes around
* hpk wonders how many people in the US are into logic programming (that will come to pycon)
<arigo> I know that both Holger and Christian are considering hiring people and giving socket as work :-)
<aleale> but auc and I could try make it mmore concrete
<hpk> arigo: well, that is quite vague still
<hpk> arigo: shouldn't stop us from tackling it at the pycon sprint in any case :)
<arigo> aleale: ok -- otherwise just mention logic and Oz in a short line
<arigo> hpk: ok
<auc> arigo: the answer depends on the state of integration of computation spaces into pypy
<auc> we don't know that in advance
* hpk could try to write a draft announcement until saturday morning, cannot promise to do it earlier
<auc> and I have no clue, currently, abouut how to do it
<arigo> ok, then to summarize the topics:
<mwh> i think at this point we want to nominate one or two people to write the announcement, not try to write it now?
<arigo> * py lib
<arigo> * gc
<arigo> * rctypes
<arigo> * logic
<arigo> * (probably a note about jit)
<auc> arigo: can you say "constraints" instead of "logic" ?
<hpk> * experimenting with pypy possibilities
<auc> 'cause constraints will be there before logic ...
<arigo> hpk: ok
<hpk> especially co-routines, thunk+X spaces etc.
<mwh> "experimenting with pypy possibilities" seems almost limitlessly vague
<hpk> mwh: we are on IRC here, aren'T we? :)
<arigo> hpk: I guess you mean in general playing with mostly app-level code that uses the new pypy features
<auc> a more focused idea could be : putting dataflow variables into pypy
<hpk> but also refining/extending the thunk space
<auc> making dataflow vars. work with microthreads
<arigo> hpk: you can already do a lot with the thunk space, I'm not sure what non-app-level extensions you have in mind
<auc> that's a building block for comp. spaces
<arigo> auc: that's more a design topic so far, isn't it?
<auc> arigo: uh ... what exactly ? (is more a design topic) ?
<auc> df vars ?
<arigo> yes, particularly how to fit them in the Python language, and how to hook them on microthreads
<arigo> unless you have more precise ideas already, of course
<cfbolz> (which we would then like to know :-)
<hpk> arigo: am mostly thinking about distribution of objects including their functions (so that not only object state is transparently moving between servers but also the code ...)
<auc> arigo: ok
<hpk> arigo: maybe that's possible already - i am not sure
<cfbolz> could we maybe try to also discuss the next topic a bit too? jan is mostly here for that
<arigo> hpk: that's also a design topic :-) I don't see clearly if and how to fit this in the thunk space in particular
<arigo> cfbolz: sure
<hpk> mwh: do we still assign to the task of finalizing the announcement? Or do you do a draft with armin or Samuele or what?
<mwh> hpk: dunno! :)
<mwh> i should be around this afternoon to work on a draft
<hpk> arigo: if you write a draft i am going to review/amend it
<arigo> let's talk about it on #pypy
<arigo> (hpk: I have some more to say to you)
* hpk will be out after the sync meeting but ok
<hpk> arigo: ok, then
<arigo> anything you want to complain about py.test?
<hpk> so any complaints about py.test?
<hpk> none? fine then we can close the meeting :)
<cfbolz> I want some features :-)
<aleale> I have had the need to be able to filter in generative test. I havent found a way to name the generative tests so that -k works.
<mwh> my pypy feature resuest: i'd like a --tb=foo thing that just listed the failing tests
<hpk> aleale: indeed, that's not possible currently i think
<cfbolz> I think -k needs to be refined in general
<hpk> mwh: no tracebacks at all then?
<mwh> in general, as arigo said we produce far too much output
<mwh> hpk: yes
<hpk> cfbolz: yes
<cfbolz> because you also cannot match the class names
<mwh> hpk: maybe the failing exception name or something
<hpk> and the file/line number i guess
<mwh> hpk: this is usually so i can work out what -k to pass :)
<hpk> makes sense i think
<mwh> yes, that might be good to work out if all the failures are like to be the same
<cfbolz> I also really want some sort of test coverage
<mwh> like -> likely
<hpk> nobody wants doctests? :)
<mwh> the "too much output" thing is partially our fault i guess
<cfbolz> and some way to add (dynamic?) tags to tests
<ericvrp> When I had some RPython code that would not annotate I would have liked a feature that simulates manual --pdb option and then looking for a translator object so t.view() can be done to see the offending block
<cfbolz> and select tests in different ways
<mwh> some kind of py.log magic might help
* hpk sidenotes that py.log is going to be refined soon - but we will ensure that pypy's usage will be fixed accordingly
<arigo> so we have mostly
<arigo> * too much output
<arigo> * better test selection
<cfbolz> * test coverage
<hpk> * doctests (i really think it makes sense also for pypy)
<mwh> hpk: i'm glad i never worked out how to use py.log then :)
<arigo> I don't feel the need for the last two items at all but I'm not against them either :-)
<cfbolz> arigo: thank you :-)
<arigo> can we close this meeting for now?
<hpk> from my side: yes and thanks!
<pedronis> I thin doctests are important form a community POV, lots of people seems into them
<cfbolz> pedronis: indeed
<mwh> yes, all from me
* arigo closes the meeting -- thanks
<hpk> * being able to run compliance/other tests with pypy-c and py.test :)
<cfbolz> bye all!
<adim> see you
<-- adim (firstname.lastname@example.org) has left #pypy-sync