Re: cpplib: Suppress warnings in system headers

[Zack Weinberg <zack at wolery dot cumb dot org>]

On Mon, Aug 21, 2000 at 08:43:31PM +0100, Neil Booth wrote:
> Zack Weinberg wrote:-
> 
> > This seems to be the right general idea.  I have one concern:  I
> > believe there are some cpp_pedwarn() calls that are not predicated on
> > CPP_PEDANTIC.  These are mandatory warnings that happen to be also
> > mandated by ISO C (so we want -pedantic-errors to make them into
> > errors).  You need to find these and give them a way to do what they
> > need. (cpp_mandatory_pedwarn?  I don't mind the long name, these should
> > be rare.)
> 
> I don't fully see what you mean. 

Your patch causes all cpp_pedwarn() calls to be silent, everywhere, if
CPP_OPTION (pfile, pedantic) is false.  This means you don't have to
test CPP_PEDANTIC() before calling cpp_pedwarn(), which is a good
idea.

But there were calls to cpp_pedwarn that didn't test CPP_PEDANTIC at
all.  For example, cppexp.c:parse_number:

  if (base <= largest_digit)
    cpp_pedwarn (pfile, "integer constant contains digits beyond the radix");

  if (overflow)
    cpp_pedwarn (pfile, "integer constant out of range");

Those pedwarns happen whether or not -pedantic is given on the command
line.  If -pedantic-errors is given, they are errors instead of
warnings, which is why they don't use plain cpp_warning().

With your patch, those warnings would happen only with -pedantic.  I
don't think that's an appropriate change.

> If what you're suggesting requires adding another category (I hope it
> doesn't), can we at the same time move to
> 
>   cpp_diag ();
>   cpp_diag_with_line ();
>   cpp_diag_with_file_and_line ();
> 
> where each takes an argument indicating severity, and scrap the
> cpp_pedwarn, cpp_warning etc.  Otherwise we start to swim in a sea of
> endless variations.

Rather than doing this ourselves, I'd prefer we look at using
diagnostic.c instead.  I think that's feasible with GDR's recent
changes, and it's where we want to go in the long run.  I'm not
opposed to this change to the *interface*, but we should be trying to
reduce the number of error reporting routines overall, not just in
cpplib.

zw