Perl 6 - the future is here, just unevenly distributed

IRC log for #pdl, 2015-08-28

| Channels | #pdl index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
11:55 opkick [pdl] wchristian pushed 3 new commits to longlong-double-fix: http://git.io/vGqWY
11:55 opkick pdl/longlong-double-fix 0f640d5 kmx: improved pdltype check
11:55 opkick pdl/longlong-double-fix 9dd507d kmx: adding test for the original SF bug #343
11:55 opkick pdl/longlong-double-fix bb6fd4a kmx: polishing doc & comments
12:00 opkick [pdl] wchristian pushed 1 new commit to longlong-double-fix: http://git.io/vGqlb
12:00 opkick pdl/longlong-double-fix 29e84a5 kmx: SF bug #343 tests in t/ops.t should run only when ivsize == 8
12:10 opkick [pdl] wchristian pushed 1 new commit to longlong-double-fix: http://git.io/vGqRR
12:10 opkick pdl/longlong-double-fix 7f0b8a9 kmx: whitespace polishing
13:17 plicease joined #pdl
13:36 chm joined #pdl
13:39 chm mohawk, sivoais:  Have you had a chance to try/review longlong-double-fix?
13:40 chm kmx has made a lot of progress and I don't have a 64bit system to test with at the moment.
13:40 chm So far all tests are passing but I don't know how many large piddle tests we have.
13:41 chm At the least, things should work with piddles >2**32 elements
13:42 chm Slicing and dicing too.
13:42 chm mapflex for large files is a good test: can we mmap a video file?
13:59 sivoais I have some large floating point volumes that I can test
14:02 sivoais I haven't looked at it yet. But I will
14:06 opkick [pdl] zmughal opened pull request #145: longlong double fix (master...longlong-double-fix) http://git.io/vGmC6
14:08 opkick [pdl] zmughal commented on issue #144: Closing in favour of #145. This will automatically test every commit made on the branch. http://git.io/vGmlB
14:08 opkick [pdl] zmughal closed pull request #144: Travis-CI (smoke longlong-double-fix) (master...lldd-travis2) http://git.io/vsxfU
14:28 chm sivoais, mohawk: Any way to have issue links like #144 be correct at http://irclog.perlgeek.de/pdl/2015-08-28
14:29 chm They seem to be set to rt.perl.org rather than our github ones
14:30 jberger chm: that's a problem with the logger system
14:30 jberger moritz might have some way to configure it, but I wouldn't know how
14:30 chm Just an annoyance but not critical.
14:32 sivoais yeah, I noticed that before. That would be nice
14:32 jberger I just asked moritz in a pm
14:32 sivoais jberger++ :-D
14:32 jberger the logger was built for logging f#perl6
14:32 jberger so my assumption is that the defaults are set for that
14:33 sivoais it can be changed in the code itself <https://github.com/moritz/ilbot/blob/ed050b334b566a7c02dee1ce5bf413876aea0321/lib/Ilbot/Frontend/TextFilter.pm#L198>
14:35 jberger oh, hmmmm
14:35 jberger so the problem is that there is no way to do metadata for a channel
14:35 jberger and so it has to be hard-coded into the logger
14:36 chm sivoais: Thanks for participating in the "Faster PDL Development Cycle"
14:36 chm sivoais: I think the PDLA/PDL::Agile is the way to go...
14:37 chm sivoais: Essentially. the only safe chage to the main PDL distribution is no direct change
14:37 chm mohawk: I'm looking forward to your thoughts
14:37 jberger chm: since you are here can I just ask, why not pull the guts to a new dist called PDL::Tiny or some such and then have PDL depend on it?
14:37 jberger I thought that had been the plan for quite a while now
14:39 chm jberger:  Mainly because of my concern that PDL would stop "just working"
14:39 jberger why should that be true?
14:39 chm Because there is a difference between official releases on CPAN vs devel.
14:40 chm If you make an "oops" then it can take perl/PDL expertise to fix thing.
14:40 jberger so you don't release an official PDL that depends on PDL::Tiny until PDL::Tiny works
14:40 chm That is why PDL-2.012 is out instead of PDL-2.008
14:41 chm Yes, that is what sivoais and mohawk have been working for.
14:42 sivoais it's a different kind of development model
14:42 sivoais I was looking at COS again yesterday and noticed these slides <https://github.com/CObjectSystem/COS/blob/master/doc/slides-design.pdf>
14:42 jberger but forking PDL into PDL and PDLA is IMO going to just lead to no adoption of PDLA
14:42 sivoais the slide on pg. 5 does a very good job of explaining the difference
14:43 chm That said, I expect we'll have some disruption and the pdl-devel discussion is to let the PDL devels add their thought and to let them know the plan going forward.
14:43 sivoais that's a very good set of slides, btw. Quite comprehensive overview of software engineering.
14:43 chm PDL::Alias is for the aggressive/agile development.
14:44 chm The stuff that works will be migrated into PDL
14:44 jberger chm: with any intention of getting merged back into PDL?
14:44 jberger ok
14:45 chm I think something like PDL::Tiny may end up being a way to use core PDL mechanics without junk
14:46 chm I'm thinking maybe (((data)types)threading)acccess) all layered on with roles as needed or desired
14:47 jberger all I wanted was a way to depend on a PDL that didn't need external libs
14:47 chm PDL::Alias will be usable but for "Get PDL" it will  be thwe main distribution
14:48 sivoais what mohawk and I were thinking was to have a PDL::Core which would be everything in Basic/ more or less
14:48 chm With plugin type framework, we could even have a pure perl PDL baseline.
14:50 sivoais chm: did you take a look at those slides?
14:51 chm Another feature of this layered approach is it should make interoperability with python/numpy and R easier
14:52 chm jberger: Have you looked at the C Object System (COS) at http://ldeniau.web.cern.ch/ldeniau/
14:53 chm plicease: ditto for you
14:53 jberger I had seen some other system lately actually
14:53 jberger I wonder if I can find it again
14:53 jberger oh right
14:53 jberger https://metacpan.org/pod/Clownfish
14:53 jberger its the basis for the Lucy module
14:53 sivoais yeah, that's the one used by Lucy
14:54 chm I'm not set on this one specifically, just not wanting to reinvent the wheel---but poorly
14:54 jberger I have no idea which is better, I just happened to see that
14:55 chm The idea is that PDL would be symmetric between the C layer and the perl layer
14:55 sivoais Clownfish is designed to make wrapping C code with a dynamic language's object system easy. It supports Perl and Ruby binding generation
14:56 sivoais though I'm not sure if it usable in a standalone fashion
14:56 chm The API is still in flux (at least that is what I think cloaked means)
14:57 chm Given we may want to keep which C OO layer we use open with some generic requirements.
14:57 jberger yeah, I don't see an independent repo for it either
14:58 jberger so perhaps its not ready for prime-time in that way yet
14:58 chm The enlightenment object model is another but is still in API flux
14:58 jberger btw
14:59 jberger Celestine Marie Mertens born on Tuesday, August 25, at 5:14 am, 7 lbs 7 oz and 21 inches long
14:59 sivoais we might have different requirements than most OO systems given we are trying to get it working with HPC
14:59 chm Since the COS is the result of some years research, I though it might be more solid.
14:59 jberger news from run4flat
15:00 chm run4flat: congrats!
15:00 sivoais jberger: great news! Send him my congratulations! :-)
15:01 jberger I will, though you can say so too the next time he pops around
15:01 chm sivoais: HPC use is another reason for the C-OO rather than C++ or others
15:02 chm We geteasy FFI access across many environments.
15:02 chm Replace TinyCC with CUDA/LLVM et voila!
15:03 chm Of course, a clean API for code generation is important to keep it simple
15:03 sivoais I will need to do some experiments and benchmarking. chm, what about starting a new GH org tentatively called PDLng for working on that? I've made one and I'll add you
15:04 sivoais and jberger, mohawk, whoever else would like to participate
15:06 sivoais added
15:06 sivoais no repos yet, but here it is <https://github.com/PDLPorters-PDLng>
15:06 jberger I'll join, but I have no idea if I'll have anything to contribute
15:07 jberger I've been trying to find some time to contribute to p5-mop-again-wtf
15:08 sivoais I think the first step should be doing a code-based "literature review" of existing libraries both for numerical computing, but also applications of it
15:09 chm sivoais: I missed the COS link but I have both his papers and the slides.  Stepped way and didn't see your line above.
15:09 chm sivoais: PDLng sounds good.
15:10 chm I'm excited about finally getting to the split and any input from PDL devels.
15:10 chm I think there may be some fun, low hanging fruit once we get started!
15:11 chm Once mohawk has time to weigh in to the "Faster PDL Devel.." thread I'll ping the other current developers directly.
15:12 sivoais I have to go for now. Trying to figure out some silly linking bug with ITK
15:12 chm Got to go for a while---will catch up later o/
15:12 sivoais o/
15:37 jberger \o
22:11 mohawk jberger, my congrats to run4flat!

| Channels | #pdl index | Today | | Search | Google Search | Plain-Text | summary