Perl 6 - the future is here, just unevenly distributed

IRC log for #marpa, 2014-06-19

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

All times shown according to UTC.

Time Nick Message
05:46 hobbs joined #marpa
11:57 LLamaRider joined #marpa
11:58 LLamaRider Hi. can someone remind me where I can find installation instructions for building marpa from source?
11:58 LLamaRider Right now I'm doing './configure && make' at ./libmarpa_build and then trying to run the main Build sequence
11:59 LLamaRider not sure if I'm missing something
12:05 LLamaRider ah, I found the make install target :)
12:38 LLamaRider sadly, I end up stuck, the cluster I am trying to install on has software that's way too old. Even autoconf is only 2.58
15:44 jeffreykegler joined #marpa
15:48 * jeffreykegler was pleasantly surprised on waking up to learn there are a lot more modules depending on Marpa than he had realized: https://groups.google.com/forum/#!topic/marpa-parser/FjreWXTQY3I
18:10 jeffreykegler joined #marpa
18:10 hobbs joined #marpa
18:10 shadowpaste joined #marpa
18:10 lucs joined #marpa
18:10 sivoais joined #marpa
18:16 jeffreykegler joined #marpa
18:16 hobbs joined #marpa
18:16 shadowpaste joined #marpa
18:16 lucs joined #marpa
18:16 sivoais joined #marpa
18:27 jdurand_ joined #marpa
19:26 hobbs I'm reading through some of the improvements in Marpa in 2014, and updating my talk, and I have to wonder, could the machinery for LATM also support "AATM" (the brute-force lexing technique from my 2013 talk)?
19:32 jeffreykegler hobbs: I'm not sure what you mean.  What slide did you talk about AATM on?
19:33 hobbs I didn't call it anything then, but slides 10-12
19:34 hobbs it's just trying all tokens that are acceptable to the grammar, and passing all that match back up as alternatives, without any kind of longest-first rule at all
19:34 hobbs which is crazy, but... :)
19:35 jeffreykegler Actually I've thought about that
19:36 jeffreykegler It's quite possible, but results in an Earley set per character, which for long inputs gets expensive.
19:37 hobbs right
19:37 jeffreykegler You can implement it in either LATM or LTM by making sure that all lexemes are exactly one character long ...
19:37 jeffreykegler and that there are no discard rules.
19:37 jeffreykegler You then need to handle whitespace in G1, which can be done.
19:38 jeffreykegler But I've found, in addition to the efficiency question, ...
19:39 jeffreykegler that the multi-layer approach seems to be natural, almost bolted into, any language people want to write and read.
19:39 jeffreykegler So I moved my experimentation elsewhere.
19:39 jeffreykegler That's not to say that someone more persistent won't find something there.
19:40 jeffreykegler Btw, this AATM (if I understand right) is in the literature, called "scannerless parsing".
19:41 jeffreykegler The SLIF's name comes from the fact that it came out of my experiments with Perl 6 and scannerless parsing.  (Perl 6 grammars are a left parsing approach to scannerless, although I'm not sure Larry would recognize them from this description.)
19:42 hobbs yeah, that's fair, actually I'm happy with LATM
19:42 hobbs I'm just thinking about the narrative of my talk while I rewrite it
19:42 hobbs and there's definitely a "push *all* ambiguity up to the top-level grammar and let it deal with it" aspect to what I was saying :)
19:43 jeffreykegler Since I really didn't go fully scannerless, I considered not calling the SLIF that, but I'd started using the term.  And it *did* dispense with the need for a Marpa-external scanner, so I called it the ScanLess InterFace or SLIF.
19:44 hobbs you'll probably have noticed by now, I'm not much of an expert on parsing, more of an interested outsider :)
19:45 jeffreykegler hobbs: re http://irclog.perlgeek.de/marpa/2014-06-19#i_8902085 -- this is an approach that Marpa makes really possible for the first time, and someone may be able to do something really amazing with it.
19:45 jeffreykegler hobbs: you and your work been one of the major influences on the SLIF, so that makes you an expert. :-)
19:46 hobbs yeah, I'm happy to see that
19:46 hobbs it was mostly accidental :)
19:49 hobbs but the usability of Marpa has come a long long way in a few years, in my opinion
19:49 jeffreykegler Btw, dunno if you caught, but for a beginning audience, you probably just want to assume an LATM world, introducing the necessary directive as a kind of boilerplace.
19:49 hobbs it's no scarier than something like Parse::RecDescent
19:49 hobbs stuff like ANTLR is way too intimidating :)
19:49 jeffreykegler And PRD and ANTLR parse a smaller set of grammars.
19:50 jeffreykegler A major advantage of Marpa is that you specify any "reasonable" BNF and you get a *linear time* parser.
19:50 hobbs yeah
19:50 hobbs vs. PRD which gets into backtracking hell very easily
19:51 jeffreykegler hobbs: exactly.  Marpa will *never* backtrack.
19:53 * jeffreykegler by "reaonable" in the above means unambiguous and no unmarked middle recursions; "reasonable" also allows various limited ambigiuities.
19:54 hobbs I only have a limited time to rewrite my talk in, so I won't be able to turn it into a very good "intro to Marpa", but I want to get across the message that it's cool, powerful, fast, actively developed, etc., and show enough sample code to give the idea that it's also approachable
19:55 hobbs I'm still shuffling things around every evening
19:55 hobbs only have a few evenings left, my talk is Tuesday before lunch
19:56 jeffreykegler hobbs: Absolutely correct.  No point in immersing them into a tutorial before they get the basic idea and realize that it's worth paying attention to.
19:56 jeffreykegler Good luck!
19:56 hobbs thank you

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