Re: [HACKERS] Postgres' lexer

> > I added the <xm> exclusive state to accomodate the possibility of a
> > unary minus. The change was provoked by Vadim's addition of CREATE
> > SEQUENCE, which should allow negative numbers for some arguments. But
> > this just uncovered the tip of the general problem...
> It seems awfully hard and dangerous to try to identify unary minus in
> the lexer.  The grammar at least has enough knowledge to recognize that
> a minus *is* unary and not binary.  Looking into gram.y, I find that the
> CREATE SEQUENCE productions handle collapsing unary minus all by
> themselves!  So in that particular case, there is still no need for the
> lexer to do it.  AFAICT in a quick look through gram.y, there are no
> places where unary minus is recognized that gram.y won't try to collapse
> it.
> In short, I still think that the whole mess ought to come out of the
> lexer...

My recollection of the whole point is that, as you mention, *you can't
identify a unary minus in the lexer*. So the minus sign is kept
distinct, to be reconciled later as either a unary minus *or* an
operator *or* whatever. The problem was that before, things like (-2)
and (- 2) were handled differently just because the spacing was
different.

Anyway, I'll look at the defacto changes; perhaps they are just fine
but I'm worried that we've reverted the behavior...
                     - Thomas

-- 
Thomas Lockhart                lockhart@alumni.caltech.edu
South Pasadena, California