Re: HEAD build failure with OpenBSD yacc

[Lennart Jablonka]

Quoth G. Branden Robinson:
> > Yes, but do look at yacc’s output, eqn.cpp.  Using OpenBSD’s yacc, eqn.cpp
> > starts with this:
> >
> >         #include <stdlib.h>
> >         #include <string.h>
> >         #define YYBYACC 1
> >         #define YYMAJOR 1
> >         #define YYMINOR 9
> >         #define YYLEX yylex()
> >         #define YYEMPTY -1
> >         #define yyclearin (yychar=(YYEMPTY))
> >         #define yyerrok (yyerrflag=0)
> >         #define YYRECOVERING() (yyerrflag!=0)
> >         #define YYPREFIX "yy"
> >         #line 19 "src/preproc/eqn/eqn.ypp"
> >         #ifdef HAVE_CONFIG_H
> >         #include <config.h>
> >         #endif
> >
> >         #include <stdio.h>
> >         #include <string.h>
> >         #include <stdlib.h>
> >
> >         #include "lib.h"
> >         #include "box.h"
> >         extern int non_empty_flag;
> >         int yylex();
> >         void yyerror(const char *);
>
> Okay, yes, I can confirm this output and the failure.
>
> It appears to me that byacc is just not playing nicely with Autoconf
> here.  And maybe not with other tools, since it's _assuming_ that the
> output will require stdlib.h and string.h.  (Its own boilerplate code
> indeed might, but its aggression here has clear negative consequences.)

PerhapsThe alternative to yacc adding #include directives is yacc 
declaring the functions it uses by itself.  It would be a little 
harder to have the yacc output not infringe on the user namespace 
in any way: If the parser wants to use memset and the user 
snippets want to define a macro called memset, what do you do?  
But I think that problem is a lot less imortant than what 
I perceive to be the actual problem:

Gnulib doesn’t play nice with others here, in prohibiting you from 
including standard library headers without doing something 
gnulib-specific, without including config.h.  I think yacc is 
entitled to believe that its output is able to include standard 
library headers.

> I don't see a path forward here except documenting that OpenBSD's byacc
> is not usable for repo builds.  (Since we ship eqn.[ch]pp in
> distribution archives, the impact on most users sourcing their groff
> from GNU should be small.)
>
> What do you think?

I think the gnulib headers should be fixed, somehow.  I don’t know 
gnulib well.  I also don’t know autoconf, automake, or bison well.

The gnulib headers could just include config.h directly.  I don’t 
know.  There may be more platforms with a yacc behaving similar to 
OpenBSD’s yacc.  And I would like to be able to uninstall bison 
again.  But I don’t care all that much.