[U-Boot] debug warning

[Albert ARIBAUD <albert.u.boot@aribaud.net>]

Hello York,

On Wed, 15 Jul 2015 13:32:30 -0700, York Sun <yorksun@freescale.com>
wrote:
> 
> 
> On 07/15/2015 01:29 PM, Simon Glass wrote:
> > Hi York,
> > 
> > On 15 July 2015 at 14:25, York Sun <yorksun@freescale.com> wrote:
> >>
> >>
> >> On 07/15/2015 01:23 PM, Albert ARIBAUD wrote:
> >>> Hello York,
> >>>
> >>> On Wed, 15 Jul 2015 13:02:07 -0700, York Sun <yorksun@freescale.com>
> >>> wrote:
> >>>> Simon,
> >>>>
> >>>> Did it happen to you with this warning?
> >>>>
> >>>> lib/fdtdec.c:108:4: warning: format ?%x? expects argument of type ?unsigned
> >>>> int?, but argument 3 has type ?fdt_size_t? [-Wformat=]
> >>>>     debug("addr=%08lx, size=%08x\n",
> >>>>     ^
> >>>>
> >>>> I think when we have 64-bit physical address, as defined in fdtdec.h, this debug
> >>>> statement needs to be changed. I am thinking to change the typedef fdt_addr_t to
> >>>> phys_addr_t, and fdt_size_t to phys_size_t. What do you say?
> >>>
> >>> I say there is no reason to change a type just because a printf format
> >>> specifier is wrong for it when building for 64-bit.
> >>>
> >>> Is there a rationale apart from the format specifier error?
> >>>
> >>> If not, then What should be done is fix the specifier so that it is
> >>> correct in both 32 and 64 bits.
> >>>
> >>
> >> Albert,
> >>
> >> Let me send a RFC patch so you can comment on it.
> > 
> > I'm not sure how to do what Albert is suggesting.
> > 
> 
> I only thought of two ways, using #ifdef or changing the type as my RFC patch
> suggests.

There's a third one: changing the form of the format specifier
according to the target's definition of the variable -- you do that
once somewhere in a common header file and you're all set up everywhere
else.

See for instance how 'uint64_t' has such a format specifier PRIu64
#define-d in /usr/include/inttypes.h; whatever the definition of
uint64_t, you can do a printf with it as follows:

	uint64_t var;
	...
	printf("Value of var is " PRIu64 ".\n", var);

(I learned this about a year ago while working on code that had to run
on both a 64-bit Intel-core PC and a 32-bit ARM-core Beaglebone)

The same technique can be used here and I strongly prefer it oer
changing the type of the variable, because changing the type affects
the function of *all* code that uses the variable (thus potentially
modifying the board behavior), whereas changing the format specifier
affects *only* the printf() code (thus not changing the board behavior
except for debug code which, supposedly, it only *fixes*.

> York

Amicalement,
-- 
Albert.