Perl 6 - the future is here, just unevenly distributed

IRC log for #pdl, 2014-06-28

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

All times shown according to UTC.

Time Nick Message
01:50 mohawk joined #pdl
01:52 Bender1 joined #pdl
01:55 mohawk joined #pdl
03:51 pdurbin joined #pdl
04:00 estrabd joined #pdl
04:02 estrabd joined #pdl
04:04 estrabd joined #pdl
09:23 sivoais for those not on the PDL list, I just posted a "status report" on the IPerl kernel <http://nbviewer.ipython.org/gis​t/zmughal/d8a37222c814956aebb8>
11:38 jberger sivoais++
12:31 estrabd YAPC::NA Science and Perl panel video - https://www.youtube.com/watch?v=tkrwAVZZgMY
13:01 mohawk joined #pdl
18:53 chm joined #pdl
18:56 chm I found the YAPC::NA Science and Perl Panel a bit disappointing (and somewhat disheartening)
18:57 chm Perhaps the issue was that generically, a perl meeting isn't where you get perl + scien(ce|tist) discussion.
18:58 chm Case in point, one of panelists was a Matlab/NumPy user pretty much unfamiliar with current PDL.
19:00 chm It did point out that we've been doing a poor job communicating to the *perl* community about *PDL*
19:01 chm We're going to see some bad press if the idea YAPC::NA attendees learned everything they need about PDL from that session.
19:01 chm jberger: As a panel member, what were your impressions?  Where do we go from here?
19:12 mohawk why did they have a matlap/numpy person on the panel?
19:12 mohawk (lab)
19:14 sivoais many different views is good, but I agree... PDL didn't get enough credit for what it does
19:15 sivoais PDL needs more documentation and blogs. That's really it. (well volunteered)  :-P
19:15 mohawk i think some solid data showing performance couldn't do any harm
19:16 mohawk but if anyone knows a real-life marketing person who could weigh in, that couldn't do any harm either
19:17 sivoais I think that David Farrell has been doing a good job with <http://perltricks.com/>
19:18 sivoais I'm currently seeing if IPerl will install on an old WinXP box, but I'm pretty sure I'd have to find the native libs.. :-(
19:19 mohawk sivoais, you're right - that is a good site
19:19 mohawk we should make cross-compiling easier
19:19 mohawk we're perl
19:19 mohawk pathologically eclectic is the name of the game
19:19 mohawk you should be able to cross-compile that thing with a snap of the fingers
19:20 sivoais hehe... I have a dream of building a similar Perl Science/Data blog called Eclectic Extractions. I just haven't had time.
19:21 sivoais well... actually, I've been spending my time building Perl tools, so I guess I'm working on it...ish
19:22 chm sivoais: More documentation and blogs visibility would be good.  I admit that my focus has been on getting PDL to build and run "everywhere"
19:23 chm mohawk: Performance data could help but I'm not sure how much of a sell to pure-perl developers
19:27 chm FWIW, I did ask waltman (the NumPy/SciPy panel member) if there was anything he thought PDL could use?
19:27 chm He didn't provide any suggestions but did admit that he was not familiar with PDL.
19:28 chm ???
19:28 sivoais what I got out of the panel was that there aren't enough complete Perl bindings to common library
19:28 sivoais *libraries
19:29 sivoais like OpenCV
19:29 chm Yes, very interesting since we've been working towards seamless library support via Alien and Alien2 technology.
19:30 chm Very surprising that no-one mentioned the Alien / Alien2 approach.
19:30 mohawk or FFI?
19:30 chm Joel was the developer of Alien::Base and he has hand-rolled some GSL bindings.
19:30 mohawk i have a vision thing that we could use Inline's glue capability to auto-interface to FFI instead of making XS
19:32 sivoais the talk on FFI at YAPC was very interesting
19:32 mohawk it really was
19:33 mohawk and you've just reminded me who plicease is
19:33 sivoais installing IPerl deps on Windows fails because no easy way to get libzmq3 and libmagic...just tested it. Those are the only problems.
19:33 chm I heard Graham's previous but similar talk at the DCBPW 2014 this past May.
19:33 mohawk i google-stumbled (googumbled) over another of his modules since then, and recognised the name but couldn't remember who he was
19:33 mohawk now i do
19:33 mohawk sivoais, nice work
19:34 chm The nice thing about a FFI approach is that we can avoid a lot of hand-XS coding for many interface cases.
19:34 mohawk yes
19:34 mohawk and if i can make my "vision thing" work (shouldn't be a massive undertaking) we can get lots of it for little effort
19:34 chm I would like to see PDL support a continuum of JIT compilation:
19:35 mohawk do we know how closely FFI works with typemaps yet?
19:35 chm gcc+XS install, gcc+Inline (semi-dynamic), C::Blocks (runtime JIT),..
19:35 mohawk (as in, is there a facility in FFI that can make use of typemaps?)
19:36 mohawk tim toady
19:41 chm It has been a while since I looked at FFI.  One possible issue (typical for Perl5) is Tim Toady (There can be more than one!)
19:42 chm sivoais: It was greatly encouraging to see your work with iPerl
19:43 chm David's new My
19:43 chm David's new yX approach to code documentation is another bright spot
19:47 chm sivoais: I would like to see OpenCV bindings as well.  The OpenCV developers have gone with hand-rolled python support
19:48 chm rather than portability so we'll have to roll our own in some form.  I was thinking a possible approach would be to use the Python
19:48 chm bindings generation and code to produce an equivalent PDL/perl one.
19:49 chm Regarding portability and generic perl development, I wonder if we could figure out a way to provide
19:50 chm PDL functionality for ordinary perl arrays, and arrays of arrays?
19:52 chm The ability to do pdl_apply(@c = @a + @b*4) might get the compute foot in the door for new users.
19:53 sivoais can tied arrays be used or are the function calls involved too slow?
19:56 chm Tied arrays are heavyweight but if you are doing number crunching at the native perl level, you're already *very* bogged down.  :-)
19:57 chm Is there an easy way to tie objects (behind the scenes) to piddles for computation
19:57 chm but have them act as the original objects to the user?
20:10 sivoais I can't think of a straightforward way that wouldn't lead to duplicating data. Then again, I don't know much about perlguts

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