# IRC log for #marpa, 2016-10-19

All times shown according to UTC.

Time Nick Message
04:05 ronsavage joined #marpa
11:01 Cheery idiosyncrat_: google postdot unique marpa
11:01 Cheery You had explained it before to me, perfectly. :)
11:29 Cheery I will do a precise explanation of this soon. And I have several ideas how to explain it well.
11:31 Cheery Okay. So here's the summary.
11:32 Cheery Once you have a properly working marpa parser sans joop leo optimizations..
11:32 Cheery the optimization is then straightforward.
11:34 Cheery Rule is leo-egilible if it is right-recursive.
11:36 Cheery an earley item is leo-egilible if it's quasi em.. this term was difficult to me..
11:36 Cheery But basically if earley item leads to reduction on the next shift.
11:38 Cheery these terms of egilibility are an optimization. The idea is to not create Leo silos for everything.
11:39 Cheery To ensure correctness, the rule of postdot uniqueness is important.
11:39 Cheery that is. In given column in parse chart.
11:39 Cheery in single earley set.
11:40 Cheery the earley item must be postdot-unique. x -> y*A
11:40 Cheery there must not be other item that also shifts with A.
11:41 Cheery If there is, we cannot ensure that the leo-silo is correct.
11:42 Cheery and this can be illustrated by an example, and a counter example.
12:04 Cheery It's actually much more complex than that, but it's represented in the setting where it's useful in your parser.
12:04 Cheery I know that because my parser was different before, and it didn't make sense. Now it makes perfect sense.
14:06 maybekoo2 joined #marpa
16:11 sirdancealot joined #marpa
16:48 sirdancealot joined #marpa
17:12 pczarn joined #marpa
17:20 pczarn Do you mean postdot uniqueness is proven to be important? Is there no way around it?
17:20 pczarn Oh, if I remember right, creating a new symbol is a way around it.
17:28 pczarn This year I'll probably find time to finish a few tasks in my parser. Once I make large progress I'm going to optimize right recursion. Only with a grammar rewrite from right-recursive to left-recursive rules, plus a rotation of bocage nodes to get the right parse.
17:28 pczarn Not with Leo optimization really.
17:30 pczarn If you have ideas for alternative optimizations, please tell me.
17:40 Cheery joined #marpa
19:34 idiosyncrat_ Cheery: nothing in the last month for "postdot unique marpa"
19:35 idiosyncrat_ Cheery: Oops -- didn't backlog enough -- just saw that one line