Re: Regression with printf 8-bit format macros

[Andre Vieira <Andre dot SimoesDiasVieira at foss dot arm dot com>]

Hi Ambroz,

Just to clarify as to why this change was made. The PRI*8 and SCN*8 
macro's were printing '*' before, the pattern for 32-bit types, when the 
actual size of __INT8_TYPE__ was '8 bits', i.e. _WANT_IO_C99_FORMATS was 
defined. Which we agreed at the time was the wrong behavior.

Now you are not the first to stumble onto this issue, though as FCh 
points out these are C99 types and the configurations you mentioned do 
not support these.

On 01/01/16 23:03, Ambroz Bizjak wrote:
> Hi Freddie,
>
> I hope we agree that C99 support should not be an all-or-nothing
> decision, you should be able to have some parts (like the stdint types
> which cost nothing), but not other parts like the full C99 support in
> printf. The worst part really is that the PRI*8 macros are defined in
> the mentioned configurations but they do not work. I see that on the
> other hand the SCN*8 macros are not defined when they would not work
> (see the top comment in inttypes.h).
I do agree that it is weird that for the SCN*8 macros to be guarded with 
_WANT_IO_C99_FORMATS, whilst the PRI*8 are not. Though guarding PRI*8 
would not solve your problem, as you seem to want these to be defined as 
'*', i.e. defaulting to their earlier behavior of returning the 32-bit 
format pattern. I would oppose doing this in all cases, since if 
_WANT_IO_C99_FORMATS is defined, then this would yield the erroneous 
behavior this patch was meant to fix.

There might be a case to be made for doing so if _WANT_IO_C99_FORMATS is 
not defined. On the other hand, doing this for SCN*8 would be a bad idea 
in my opinion and this would keep the discrepancy between PRI*8 and SCN*8.

I do not think we should encourage the use of PRI*8 and SCN*8 types if 
these are not supported by the library being used. As this seems to be 
the case, since _WANT_IO_C99_FORMATS is not defined, which I think would 
be the appropriate macro to indicate support for such.

> But then we can have the PRI*8 macros for free anyway, so why should
> they be available only when printf has "hh" support?
>
> Best regards,
> Ambroz
>
>
> On Fri, Jan 1, 2016 at 11:44 PM, Freddie Chopin <freddie_chopin@op.pl> wrote:
> > On piÄtek, 1 stycznia 2016 12:08:30 CET Ambroz Bizjak wrote:
> > > Hello,
> > >
> > > It appears to me that commit "Fix for pri and scn formats" [1] broke
> > > the 8-bit printf format macros (PRI*8), for scenarios when newlib is
> > > configured either with --enable-newlib-nano-formatted-io or without
> > > --enable-newlib-io-c99-format is not used. Because the new code
> > > defines these macros to the "hh" formats, which are not supported
> > > then.
> >
> > Is it actually a problem that PRI* macros don't work without support for C99?
> > After all, these are used for C99 types, so not really needed without C99...
> >
> > Regards,
> > FCh

Further discussion is most welcome!

Cheers,
Andre