[developers] Stack overflow in generator unpack-edge

Dan Flickinger danf at stanford.edu
Sat Jan 16 10:11:08 CET 2010

Thanks, Berthold, for that very good suggestion to run the generator on my problematic example with packing turned off.  I could then immediately see that I had recently introduced a bug into the lexical entry for the complementizer `for', which led to serious overgeneration.  Now that I have fixed the bug, the example generates fine, both without and with packing.  

Thus it seems that at least in some cases, the generator with packing on will produce a "Stack overflow" message on an item where generating without packing produces the more informative usual warning about a possibly recursive rule, which is then immediately visible in the generator chart.   So perhaps there is no need to change anything in the generator -- sensible grammar developers will of course remember to examine problematic generation examples with packing disabled.

Sorry for the false alarm, developers.


----- Original Message -----
From: "Berthold Crysmann" <crysmann at ifk.uni-bonn.de>
To: "Dan Flickinger" <danf at stanford.edu>
Cc: "developers" <developers at delph-in.net>
Sent: Friday, January 15, 2010 1:26:56 PM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna
Subject: Re: [developers] Stack overflow in generator unpack-edge

On Thu, 2010-01-14 at 23:38 -0800, Dan Flickinger wrote:
> Using the current SVN versions of the ERG and the LKB, I get a stack overflow error resulting from a tight self-calling loop in the generator's unpack-edge!!() when generating (with packing) from the expletive-it subject (NP-X) reading of the sentence 
>   It fails to rain.
> It seems that the generator may be having difficulty distinguishing the edge "to rain" from the edge "for it to rain", both of which have identical semantics.
>   Dan
Hi Dan, 

I occasionally bump into this with semantically vacuopus triggered
items.  What happens if you switch  packing off? Does "it" (or "for it")
combine more than once? 


More information about the developers mailing list