Perl 6 - the future is here, just unevenly distributed

IRC log for #pdl, 2015-10-24

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

All times shown according to UTC.

Time Nick Message
13:20 jberger joined #pdl
13:52 chm joined #pdl
17:40 piko joined #pdl
17:45 piko Hi, I am thinking of replacing Octave with PDL, I browsed some PDL pages and I am wondering, is the PDL project still alive?
17:45 mohawk piko, yes it is
17:45 mohawk there has just been a new release
17:46 mohawk if you'd like to sample what you might call a remix of the project, please take a look at PDLA (doesn't yet incorporate a few feature improvements and bug-fixes from PDL)
17:50 piko mohawk: and is it just my impression that the code may benefit from some improvements e.g. to get rid of warnings and some testing?
17:53 piko mohawk: I just ask since I saw some Perl crash in perl-5.20 in a demo
18:18 piko mohawk: I will try PDLA
18:57 chm mohawk: Hi, now that we have 64bit support in PDL, I was wondering what the status is for PDLA?
18:59 chm piko: As mohawk said, we've just finished a release adding full 64bit integer support to PDL and some bugs fixed.
19:00 chm piko: regarding "is it just my impression that the code may benefit from some improvements", yes, starting with bug reports.
19:00 chm The recent PDL release is the second in the past 2 months.
19:01 chm If you count the formation of the Agile PDL version then that is 3 major releases in the past few months.
19:02 chm piko: what version of PDL are you using, what hardware, what demo, what was the crash...
19:03 chm piko: There are some issues with the latest PDL releases and some external modules for PDL.
19:03 piko chm: that sounds good, still how big is the coder/tester base and is it likely that the architecture/interface would radically change in the near future?
19:03 chm piko: It takes time to update them with the new API and some developers are busy with non-PDL stuff/life...
19:04 piko chm: have a look at https://bugs.gentoo.org/show_bug.cgi?id=563376
19:05 piko chm: of course, no problem with life :-) I can stand some bugs and anyway I search some opensource project to get slightly involved in, I like perl and I do some engineering stuff, so this looks good to me
19:06 chm piko: I don't see the PDL version info in the bug you report.
19:06 chm piko: since gentoo often builds completely from scratch, there are a lot of ways problems could occur.
19:07 chm piko: that said, for all the platforms we have bug reports for and with someone able to test/report/try, we fix them.
19:10 chm piko: regarding architecture/API changes, the idea for PDLA is to split the original, monolithic PDL distribution into separate modules that can be developed more efficiently.
19:10 chm piko: Also it would allow PDL to become more of a PDL-Core + other modules
19:11 chm piko: PDL-Core would contain the low level compute/data types support.
19:12 piko chm: version 2.4.11 ... you may be right with build issues, I will rebuild all the perl modules to see if it gets fixed
19:13 chm piko: We're planning to implement new PDL-Core with improved architecture for modularity, more data types, greater performance and generality.
19:14 piko chm: is that plan described somewhere?
19:14 piko chm: ... or do you work more in agile way?
19:15 chm piko: There has been some discussion on the PDL mailing lists and on #pdl.
19:17 piko chm: okay ... and the best documentation resource to start with seriously - PDL book?
19:18 chm piko: I've been the PDL release manager for some versions now (17-ish)
19:18 chm piko: I plan to keep PDL-2.x in stabilized mode so we always have a version to rely on.
19:19 chm piko: In the meantime, the new capabilities and compute kernel will be applied to the PDLA
19:19 chm piko: That would be merged into the PDL releases as the technology becomes mature.
19:20 chm piko: The PDL book isn't perfect but it covers a lot of PDL capability.
19:21 chm piko: support is via our sf.net mailing lists and the development is based on github.org
19:21 chm piko: Are you familiar with perl?
19:22 piko chm: I would say slightly advanced in Perl
19:22 piko chm: still I learn new things abut Perl all the time
19:23 piko chm: like POE, Moose or even internals
19:23 chm piko: Then the PDL Book should be fine, since you already know the perl part of things.
19:25 piko chm: okay, will start there ... one more thing - using PDL as a replacement of Matlab/Octave requires some level of reliability, what level of testing do you want to have?
19:26 chm piko: Matlab/Octave has a much larger user base and lots of packages and modules available.
19:27 chm piko: PDL has some uniq capability, especially with vectorized computations (so called "threading")
19:28 piko chm: sure, and I am okay with that ... my motivation is to have computing and Perl at the same time, so I am okay with missing modules and I think I can write what I need in Perl/PDL. I more fear about bugs that could change my results without notification.
19:28 piko chm: ... looking at the threading right now at PDL::Core
19:29 chm piko: The biggest single limitation is that the PDL-2.x types don't support arrays of structs or pointers
19:29 chm piko: By re-engineering the Core we'll get much better interoperability with other tools Octave/Numpy/R
19:30 jberger piko: while the userbase is smaller, some of those users are no-fail users, large scientific projects and the like
19:30 chm piko: basically because we'll be able to read/write/grok their types which is not possible with PDL-2.x
19:31 chm piko: A number of the PDL developers have been using PDL for work/research computation.
19:31 piko jberger: no-fail means intolerant to failures?
19:31 jberger crashed satellites or something
19:31 chm piko: One of the plans for PDLA is to fold in some test based development for the new features.
19:32 chm piko: Also to add more tests for existing features so that the backport to PDL can be validated when made.
19:32 piko jberger: oh, I see
19:32 jberger chm: piko just wants you to tell him/her that its not going to break
19:33 jberger as I know that's one of your distinct goals, best to just say that
19:33 jberger :-P
19:33 chm jberger: definitely!  sometimes it is hard to figure out what is being asked or what is the right answer (out of many), thanks!
19:34 piko jberger, chm: sounds serious and ambitious, okay will start there ... if I wanted to get involved let's say from the Perl side first, is there any page listing work needed for the project that a contributor can start on?
19:34 chm jberger: I'm in the process of building gcc-5.2 so that I can try out their gcc-jit library.
19:35 jberger piko: this is the github issue tracker: https://github.com/PDLPorters/pdl/issues
19:35 chm jberger: Also, the C Object System (COS) looks pretty good for the new PDL Core engine (OO, multimethods, around methods, C for FFI...)
19:35 jberger I think most major issues have been ported from sf.net
19:36 * jberger is thankful to be doing mostly web now and not worrying about objects at the C level :-P
19:36 chm jberger, mohawk:  Have you looked at what would need to be implemented to use a new Core with the PDLA outsides?
19:37 chm jberger, mohawk:  Would we need to implement all the low level API routines or what?
19:37 jberger I wouldn't have the slightest clue, I doubt mohawk would either
19:37 chm I haven't gotten in to it myself but was wondering if you had already investigated some...
19:37 jberger run4flat was the guy for that, but I don't know if he has any more PDL time
19:38 jberger run4flat had a lot of that in his head and had big ideas about what a revamp would look like
19:38 chm jberger: no worries, just avoiding duplication of effort.  Initially I'm not going to worry for the first couple of iterations.
19:39 chm Don't want to lock in a new architecture until we've used up a couple of throw aways---so we know it's clean!
19:39 chm I've got to head out.  o/
23:49 lungching joined #pdl

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