On Mon, 2006-12-11 at 18:40 -0500, Christopher Browne wrote:
> The set of things fixed in 1.2.2 is listed below.  Important Issues
> that would warrant dropping everything to look at this are listed as
> *CRITICAL*...
> 
> I have been holding off based on the search for the problem Peter
> Eisentraut reported where if you neglect to have a '\n' at the end of
> a slon.conf file, on some platforms it'll throw an error.  That
> problem was present from probably the beginning of the 1.1 cycle, so
> it seems less than *totally* urgent to get it resolved
> instantaneously.  
> 

I didn't find Peter's original problem, so I'm not sure I understand. I
wouldn't expect this to be platform-dependent; which platforms have
which behavior?

> On that note, is there a Flex expert in the house?  We end the parser
> with a generic "match anything" rule that returns SLON_FERROR,
> consistent with the way the PostgreSQL GUC parser works.  The error
> that is thrown is based on an EOF falling into this rule.  Darcy
> Buskermolen is speculating that some variable is getting "smashed"
> along the way.  I'm just not sure; I'm a bit confused that there's not
> an explicit search for an EOF...
> 

I am no Flex expert (far from it), but I read the logic in guc-file.l to
mean that it succeeds when a line is terminated with an EOF, and the
logic in conf-file.l will error if a line is terminated with EOF.

However, guc-file.l is somewhat confusing because it doesn't test
against EOF, it tests if (token == 0). I suppose in Flex, token == 0
means EOF?

Might it work to just change (token != EOL) to be (token != EOL &&
token != 0) in "case 2:" of conf-file.l?

Regards,
	Jeff Davis