Probable bugs in egcs-19981220 and autoconf 2.12

[Manfred Hollstein]

On Tue, 22 December 1998, 21:24:25, bagnara@di.unipi.it wrote:

 > N8TM@aol.com wrote:
 > > 
 > > In a message dated 12/22/98 3:57:22 AM Pacific Standard Time,
 > > bagnara@di.unipi.it writes:
 > > 
 > > << I discovered the following while using autoconf 2.12
 > >  together with egcs-19981220/g++. Autoconf uses the following
 > >  program to check whether the C++ compiler works:
 > > 
 > >  #line 894 "configure"
 > >  #include "confdefs.h"
 > >  main(){return(0);}
 > > 
 > >  If the -Wall option is used, then egcs-19981220 rejects it >>
 > > This could be useful information.  egcs expects a newer version of autoconf,
 > > but I have had to fall back on autoconf 2.12 for systems where the newer
 > > version refuses to install.  Maybe this change and the change of the version
 > > report are all that are required.  However, this test string appears on first
 > > glance to appear in acspecific.m4.
 > 
 > Perhaps my message was not clear enough, sorry about that.
 > The bugs do not show up building egcs (the problem I reported previously
 > with the egcs configuration mechanism expecting a non-released version of
 > autoconf seems to have been fixed). I have discovered them building
 > an application of mine.
 > 
 > As far as egcs-19981220 is concerned: -Wall should not cause the rejection
 > of a C++ program just because it omits the return type of `main'.
 > IMHO, either the program is rejected regardless of -Wall, or the
 > program is compiled.

I  posted a small patch for  this with the subject ``CVS-19981219: New
PATCH for actually fixing the "main(){}" case''  some days ago (please
see the appended copy).

Jason (Merrill): I haven't heard anything from you so far, is it OK to
install?

A Happy New Year to everybody out there!

manfred

 > 
 > As far as autoconf 2.12 is concerned: testing whether the C++ compiler
 > works should be done with a program not omitting the return type of `main'.
 > 
 > The combination of these two bugs makes autoconfiguration of C++ programs
 > impossible if the user has specified CFLAGS so as to contain -Wall.
 > I am one of those users ;)
 > 
 > Cheers,
 > 
 >          Roberto

On Sat, 19 December 1998, 20:14:43, manfred@s-direktnet.de wrote:

 > On nil, 17 December 1998, 09:48:15, jason@cygnus.com wrote:
 > 
 >  > No; we want -Wreturn-type to warn about this case.  It ought to be a plain
 >  > warning, though.
 >  > 
 >  > Jason
 > 
 > OK.  How about this, then?
 > 
 > 1998-12-19  Manfred Hollstein  <manfred@s-direktnet.de>
 > 
 > 	* decl.c (user_wants_pedantic_errors): Define new variable.
 > 	(init_decl_processing): Record the current value of flag_pedantic_errors.
 > 	(grokdeclarator): Define new helper variable `is_main'; decide based on
 > 	`user_wants_pedantic_errors' if "main () { ... }" should be flagged with
 > 	an error message or with a warning only.
 > 
 > diff -rup -x CVS -x RCS -x *.o -x *.info* -x *.html* -x *.elc -x *.dvi -x *.orig -x *~ -x version.el egcs-19981219.orig/gcc/cp/decl.c egcs-19981219/gcc/cp/decl.c
 > --- egcs-19981219.orig/gcc/cp/decl.c	Sat Dec 19 10:24:37 1998
 > +++ egcs-19981219/gcc/cp/decl.c	Sat Dec 19 19:16:52 1998
 > @@ -435,6 +435,9 @@ static struct named_label_list *named_la
 >     in the TREE_PURPOSE slot.  */
 >  tree static_aggregates;
 >  
 > +/* Record if the user specified -pedantic-errors.  */
 > +static int user_wants_pedantic_errors = 0;
 > +
 >  /* -- end of C++ */
 >  
 >  /* Two expressions that are constants with value zero.
 > @@ -5830,6 +5833,10 @@ init_decl_processing ()
 >  
 >    if (flag_strict_prototype == 2)
 >      flag_strict_prototype = pedantic;
 > +  /* Record if the user gave the -pedantic-errors flag; we will need this
 > +     later to decide, if "main () { ... }" should be flagged with an error
 > +     or with a warning only.  */
 > +  user_wants_pedantic_errors = flag_pedantic_errors;
 >    if (! flag_permissive && ! pedantic)
 >      flag_pedantic_errors = 1;
 >  
 > @@ -9353,16 +9360,24 @@ grokdeclarator (declarator, declspecs, d
 >  	}
 >        else
 >  	{
 > -	  if (! pedantic && ! warn_return_type
 > -	      && funcdef_flag
 > -	      && MAIN_NAME_P (dname)
 > -	      && ctype == NULL_TREE
 > -	      && in_namespace == NULL_TREE
 > -	      && current_namespace == global_namespace)
 > +	  int is_main = funcdef_flag
 > +			&& MAIN_NAME_P (dname)
 > +			&& ctype == NULL_TREE
 > +			&& in_namespace == NULL_TREE
 > +			&& current_namespace == global_namespace;
 > +
 > +	  if (! pedantic && ! warn_return_type && is_main)
 >  	    /* Let `main () { }' slide, since it's so common.  */;
 >  	  else
 > -	    cp_pedwarn ("ANSI C++ forbids declaration `%D' with no type",
 > -			dname);
 > +	    {
 > +	      int old_flag_pedantic_errors = flag_pedantic_errors;
 > +
 > +	      if (is_main)
 > +		flag_pedantic_errors = 0;
 > +	      cp_pedwarn ("ANSI C++ forbids declaration `%D' with no type",
 > +			  dname);
 > +	      flag_pedantic_errors = old_flag_pedantic_errors;
 > +	    }
 >  	  type = integer_type_node;
 >  	}
 >      }
 >