Re: [C PATCH] Better column info for initializers (PR c/60139)

[Marek Polacek <polacek at redhat dot com>]

On Thu, May 01, 2014 at 11:55:38PM +0000, Joseph S. Myers wrote:
> On Fri, 25 Apr 2014, Marek Polacek wrote:
> 
> > Another minor fix: use loc instead of input_location.  Also add
> > missing OPT_Wpedantic. 
> 
> Why do you say it's missing?  A default pedwarn generally means "this is 
> dubious code, not allowed by ISO C, but we don't want to make it a hard 
> error" (and the trend is more to making such things hard errors, rather 
> than limiting them to -pedantic).
> 
> For the *first* such change ("not computable at load time"), I think it's 
> already conditional on (pedantic && !flag_isoc99), based on the setting of 
> require_constant_elements, and so using OPT_Wpedantic is appropriate.  But 
> I think the second represents an undesirable change to how the compiler 
> behaves - code such as
> 
> int f();
> int a = 1 ? 1 : f();
> 
> *should* be diagnosed by default rather than needing -pedantic.  If you 

Yeah, this case is even now diagnosted by default (for this we warn in
digest_init).

> think this case also only affects for form of diagnostics that were 
> already only given if pedantic, please explain further why this is the 
> case.

Right.  I added OPT_Wpedantic to second pedwarn_init ("initializer
element is not a constant expression"), because this warning was
output only when using -pedantic and C89, but it didn't say "[-Wpedantic]",
thus it looked like the warning is enabled by default.  The reason
is that this pedwarn is guarded by require_constant_elements, which
is, as you said, dependent on (pedantic && !flag_isoc99) in
start_init.  Just for the record, this pedwarn warns e.g. on

double sin (double);
void
fn (int *p)
{
  double d[] = { sin (1.0) };
}

So the patch doesn't suppress any warnings, it only adds
"[-Wpedantic]" where it should be (there might be other cases where 
"[-Wpedantic]" is missing).

	Marek