Re: [PATCH] -D, --define.

[Akim Demaille]

Le 10 nov. 08 à 15:48, Joel E. Denny a écrit :

> On Mon, 10 Nov 2008, Akim Demaille wrote:
>
> > > In muscle_tab.c and bison.m4,  
> > > `[Bison:muscle_percent_define_default]' and
> > > `[Bison:b4_percent_define_default]' are some other pseudo file  
> > > names.
> > > Unlike `<command line>', these should only be printed by Bison when
> > > there's a developer bug.  Nevertheless, we should be consistent  
> > > about
> > > whether we use square or angle brackets.  I don't care which.
> >
> > You are right.  I would then promote the <> notation, together with  
> > a human
> > readable location.  The examples you mention look way too much like  
> > a bug in
> > Bison (this is was I was told by co-workers who happened to have  
> > seen these
> > locations) because they look like something that was not expanded.
>
> I remember you reported that.  [] does look like overquoted M4, so I  
> agree
> that <> is better.
>
> Just to be clear, when these locations are printed, there is a bug in
> Bison... or at least in a skeleton.  That is, I can't imagine that  
> Bison
> should have any reason to report the locations of internal defaults  
> unless
> there is something wrong with them.  Of course, if you'd like to  
> propose
> better messages here, I'm definitely fine with that.

I was not referring to locations displayed by Bison, but of locations  
left as synclines.

address@hidden ~/src/urbi/2.0/kernel $ grep -A2 Bison: src/parser/ 
ugrammar.cc
#line 1 "[Bison:b4_percent_define_default]"

namespace yy {
--
#line 1 "[Bison:b4_percent_define_default]"

} // yy


Maybe that's a case where we should not issue the #lines at all.