New proposed #parrotsketch protocol: trac.parrot.org/parrot/wiki/Propos...chProtocol | Please prepost reports by 1800 UTC. | Logs: tinyurl.com/parrotsketch
Set by moderator on 30 June 2009.
01:37 ascent joined 01:57 ascent joined 02:09 ascent joined 03:52 ascent joined 08:56 masak joined 11:57 ascent joined 12:18 whiteknight joined 13:50 PacoLinux left 15:12 whiteknight joined
whiteknight (Preposting very early because I'm busy. Might not be at the meeting today) 15:12
What I did:
- Fixed an issue (TT#834) where exit codes were not being set correctly from die 15:13
- Added some documentation about "MRO" and "C3" per request
- Work on Windows 64-bit platform. Parrot now builds there (with some Configure.pl command line help) but doesn't pass all tests 15:14
What I will do:
- Helping with some of the deprecation tasks (if bacek++ doesn't steal them all first!) 15:15
- More IO Work, hopefully starting on AIO.
What I am blocking on:
- Nothing
EOR
15:46 particle left 15:48 particle joined 16:21 Util joined
cotto # What I did: 16:32
* got opsc (+ tests) working with explicit ops preambles
* kicked 1.4 out the door (+ associated release managering)
# What I hope to do and how many tuits I expect to have:
* now back to your regularly scheduled opsc hacking (tuit outlook is good)
.eor
17:19 allison joined
Util # Done: 17:26
* Tried TT #835, but could not find the GC bug.
# Plan for next week:
* Giving talk at Atlanta.pm on YAPC::NA::2009, including Parrot VM Workshop details.
* TT #691 Candidate patch
* Kill off Parrot_io_write
# Blockers:
* No tuits until Saturday evening.
.end
17:37 fperrad joined
fperrad # What I did : 17:37
* smoke some languages before release
* release 1.4.0 : Windows packages at SourceForge 17:38
17:38 Tene joined
fperrad .eor 17:38
allison - Worked on install tickets. Made good progress, you can now build and test Pynie and Tcl completely from an installed parrot (with a removed build directory) on various flavors of Linux. Mac OS X still has library linking issues. 18:11
- Built Debian and Ubuntu preview packages for 1.4.0, no problems.
- Will be mostly AFK until Friday when OSCON ends, but will have substantially more free time after that. My priorities are Debian/Ubuntu packaging first, then pcc_rewiring. Drop me a message or catch me on IRC next week if there are items you'd like to make sure get on my task list.
EOR
18:12 darbelo joined 18:19 jhorwitz joined
darbelo posted his #ps report at www.parrot.org/content/week-decnum-dynpmcs.-0 18:22
18:29 chromatic joined
chromatic Fixed some bugs. 18:29
Will remove some deprecated features and work on pluggable runcores this week. 18:30
Hello, everyone. 18:32
allison hi 18:33
Util hi
fperrad hello
japhb o/ 18:34
chromatic Where's our 1.5 roadmap? 18:35
allison trac.parrot.org/parrot/report/14
we also need to review 2.0 tasks to spread them across the next 6 months 18:36
cotto hi
chromatic Let's set some priorities though.
18:36 NotFound joined
chromatic Suggestions: PCC rewiring, PIRC. 18:36
allison our overall priority for the next 6 months is "production users"
chromatic That's vision.
moderator Vision for 2.0: Production Users 18:37
allison the monthly priorities should support that 18:37
NotFound Sorry, I'm clock-blind today.
allison pcc_rewiring is an optimization, and so supports the overall goal. PIRC is interface and internals cleanup, so supports the same
cotto How does pirc fit in with a pure pir pir compiler? 18:38
allison cotto: they're orthogonal
chromatic Easier to maintain than IMCC until we use only a pure PIR PIR compiler. 18:39
cotto ok
allison the two other highest priorities from 2.0 I see are "prune c data structures" which should be scheduled early, perhaps 1.6 18:40
and "documentation sprint", which should be on the priority list for every month between now and 2.0
chromatic Anyone else, thoughts? 18:41
Tene allison: pcc_rewiring isn't just an optimization. It also allows :vtable(invoke), which we can't do otherwise.
NotFound And it will allow to clean and fix other things, I hope. 18:42
allison Tene: curiously, for once the optimization itself is a priority (in addition to opening up new features and cleanups)
japhb Is the idea that people be using pure PIR code in production at 2.0, or use a Parrot HLL in production at 2.0?
Tene Okay.
allison japhb: both 18:43
japhb s/at/soon after/, if you like
Then my impression is that there's going to have to be a LOT of optimization between now and then. Not percentages faster, but multiples might be nice.
Like a similar improvement as with the IO layer.
chromatic We might be able to double the speed. 18:44
allison are there specific optimization tasks we could add to the roadmap?
(a general "optimize" task is too broad) 18:45
spinclad lots of profiling
japhb Rakudo is, depending on operation, 1-2 orders of magnitude too slow for some of the stuff I want to do. And I went spelunking, a lot of that appears to be the "Perl 6 lexical scope == PIR sub" equivalence. It's REALLY slow.
chromatic The profiling runcore is essential.
allison japhb: so this may be a good time to explore scopes smaller than a sub
japhb I have a couple loops where I replaced an inner loop of a couple lines of Perl 6 with an equivalent Q:PIR, and got 6-10 times faster. 18:46
allison japhb: more experimental, so wouldn't put it on the roadmap, but worth an RFC ticket
japhb The other problem may be a Perl 6 spec issue, or may be a Rakudo implementation issue: A lot of overhead is created by needing to make new $_, $!, $/ each scope 18:47
(I think those are the three, I don't have the PIR in front of me)
chromatic This is getting technically specific; it may be better in #parrot or #perl6.
japhb sorry
Just wanted to bring up some of the issues that Parrot *may* be able to help with to get the HLL's faster. Because right now the HLLs seem to be spending their energy just trying to get fully working, and Parrot is arguably already there and on to the next task. 18:48
chromatic Sure; this is the time to discuss development priorities. I don't want to get too deeply into how they work, just what they are. 18:49
japhb If we expect HLLs to be production-usable when Parrot reaches 2.0, Parrot needs to help them get there performance-wise.
allison japhb: yes, performance is definitely a high priority for 2.0
NotFound Some optimization in the code generation by the compiler may also be a good approach. 18:50
japhb NotFound: Yes. And I think PCT and lower layers may be a good place for that.
Language-specific code gen improvements are great, but they don't benefit the other HLLs. :-)
NotFound Sure, if PCT does some optimizations, we can hope it will do for all compilers.
japhb Is pmichaud around? 18:51
chromatic Should we consider an optimization stage for PCT?
japhb chromatic: I think that would be very valuable, yes.
allison chromatic: it has definite potential, and we've designed it to accept one
NotFound: could you make an RFC ticket for potential PCT optimizations, we can review it next week and decide whether to add it to the roadmap
NotFound allison: I don't think I'm familiarized enough with PCT for that task. 18:52
allison NotFound: you can make it a list of questions
japhb What's PCT's bus number? 2? (pmichaud and Jonathan?)
NotFound I'll give it a try, then. 18:53
chromatic Let's move to questions. Does anyone have a question?
Okay. Let's do a brief review of the 1.0 -> 1.4 releases. 18:55
What went right?
NotFound Quick question:
Is someone still working on longopt refactoring?
Looks like no ;) 18:56
Util NotFound: ticket number?
NotFound I think there is none, just hear someone commenting about some months ago on irc. 18:57
The issue was to wait for that before stopping to depend on imcc_main from parrot main. 18:58
Forget it, I'll open a ticket. 18:59
chromatic Thoughts on what went right? 19:00
allison It was a very smooth release, cotto++ for the advanced planning. 19:01
Util Went right: released on time and followed deprecation policy (met our promises to users)
allison We got good platform testing.
chromatic The development priorities of the last month seemed to work fairly well. 19:02
What went wrong? 19:03
japhb chromatic: Unfortunately, if priority 1 and priority 1.5 were well met, priority 2 (HLL interop) got less attention. 19:04
Not "none", but not enough.
chromatic Priority 3 had one person working up until Sunday night. 19:05
allison One branch merge last week was not as smooth as it could have been. For 2.0 we may want to consider holding off branch merges entirely in the week before the release. 19:06
chromatic We put off a lot of tickets and milestones again. 19:07
allison Yes, we're getting better at estimating, but still further to go. 19:08
chromatic I think setting priorities and finding multiple people for tasks will help.
allison I think so too 19:09
japhb Instead of milestones, why not a prioritized (longer) list? It's a bit of a downer to miss goals each release.
allison Also, setting expectations more accurately. If we divide up the 2.0 roadmap tasks we'll have 1 or 2 each month, instead of the 6+ we had every month between 1.0 and 1.4 19:10
chromatic Maybe we should prioritize them based on niceness to have, so that we'll work on the most important sooner.
allison japhb: the advantage of dividing it up monthly is you *know* what to work on that month, instead of staring at a large list and being overwhelmed
japhb: but, it is okay for tasks to slip between months, as long as they're on target for 2.0 19:11
chromatic: also, we may need to break down some of the tasks so they have multiple entries
japhb allison: do we believe that if we slip a little each month, there's enough slack to still be on target for 2.0? 19:12
allison chromatic: I noticed that we scheduled some of the bigger tasks for the end in the last cycle, with the thought that we'd have 4 months to work on it, but no entry for that work in earlier months
chromatic Sounds like we're in the "What to change?" stage.
allison chromatic: yes 19:13
chromatic Let me summarize then: 19:14
allison japhb: that depends on the slippage, if it's "no work was done at all in this month" then no. If it's "branch merged in just after release" then yes.
chromatic - schedule milestones by their importance
- break milestones into smaller pieces 19:15
- recruit multiple people to work on each milestone
allison chromatic: sounds like good action items
chromatic Any other discussion? 19:16
Util Is the plan for the PIR profiler written down anywhere, or just in key people's heads? 19:18
chromatic Putting it on the Wiki now.
Util Excellent
chromatic Let's call that a day. Thank you everyone. 19:26
19:26 Util left, NotFound left, fperrad left 19:28 darbelo left 20:13 eternaleye joined

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!