Preprocessor: some dollars-in-identifiers mishap broke cris-axis-elf

[Neil Booth]

Hans-Peter Nilsson wrote:-

> > Excess errors:
> > <built-in>:56:21: warning: '$' in identifier or number
> >
> > "everywhere" (1000:s of test-cases) for example
> > gcc.dg/20000111-1.c.  Note #define REGISTER_PREFIX "$" but
> > that's all dollar-wise.
> 
> As I guessed, it's exactly (in a callee of) the
>  builtin_define_with_value ("__REGISTER_PREFIX__", REGISTER_PREFIX, 0);
> in c-cppbuiltin.c that the warning is emitted.
> 
> I think GCC shouldn't warn on that macro definition in
> particular and perhaps not on builtin definitions in general.
> 
> Neil, what's preferred and what's the best way to accomplish it?

The warning is controlled by a flag warn_dollars.  How about refining
when this is set?

Currently cppinit.c has:

     pfile->warn_dollars = CPP_PEDANTIC (pfile) && !CPP_OPTION (pfile,
     c99);

This is a cpp-internal flag in cpp_reader.  I suggest we make it settable
by the client (i.e. cc1) by putting it in the cpp_options structure and
remove the above line, that cpplib as a library defaults to true in
cpp_create_reader, and that c_common_post_options does

     cpp_opts->warn_dollars = (CPP_PEDANTIC (pfile)
     			       && !CPP_OPTION (pfile, c99) 
			       && DOLLARS_IN_IDENTIFIERS);

or something similar, so that dollar targets only warn if both not
c99 and pedantic.  Note that cpplib must not have target dependencies,
of which DOLLARS_IN_IDENTIFIERS is one.

Since you have this issue, if you're happy with my suggestion, could
you create and test a patch along the above lines?  Such a patch is
pre-approved.

Thanks,

Neil.