On 06 Oct 2022 14:39, Tsukasa OI via Gdb-patches wrote:
> On 2022/10/05 20:45, Andrew Burgess wrote:
> > Tsukasa OI <research_trasio@irq.a4lg.com> writes:
> >> Clang generates a warning if the format parameter of a printf-like function
> >> is not a literal.  However, on hw_vabort, it's unavoidable to use non-
> >> literal as a format string (unless we make huge redesign).
> >>
> >> We have "include/diagnostics.h" to suppress certain warnings only when
> >> necessary.  Because DIAGNOSTIC_IGNORE_FORMAT_NONLITERAL can suppress
> >> warnings when the format parameter of a printf-like function is not a
> >> literal, this commit adds this (only where necessary) to suppress this
> >> error with "-Werror", the default configuration.
> >>
> >> sim/ChangeLog:
> >>
> >> 	* common/sim-hw.c (hw_vabort): Suppress non-literal printf warning
> >> 	by using DIAGNOSTIC_IGNORE_FORMAT_NONLITERAL.
> >> ---
> >>  sim/common/sim-hw.c | 3 +++
> >>  1 file changed, 3 insertions(+)
> >>
> >> diff --git a/sim/common/sim-hw.c b/sim/common/sim-hw.c
> >> index cece5638bc9..36f355d2262 100644
> >> --- a/sim/common/sim-hw.c
> >> +++ b/sim/common/sim-hw.c
> >> @@ -425,10 +425,13 @@ hw_vabort (struct hw *me,
> >>    strcat (msg, ": ");
> >>    strcat (msg, fmt);
> >>    /* report the problem */
> >> +  DIAGNOSTIC_PUSH
> >> +  DIAGNOSTIC_IGNORE_FORMAT_NONLITERAL
> >>    sim_engine_vabort (hw_system (me),
> >>  		     STATE_HW (hw_system (me))->cpu,
> >>  		     STATE_HW (hw_system (me))->cia,
> >>  		     msg, ap);
> >> +  DIAGNOSTIC_POP
> > 
> > Rather than disabling diagnostics, I'd like to propose the patch below
> > which expands FMT and AP within sim-hw.c, then passes the expanded
> > string through to sim_engine_abort.  What do you think of this?
> 
> Ah, It took a while to understand but makes sense to me.
> 
> I just needed to add ATTRIBUTE_PRINTF (2, 0) to suppress "-Werror
> -Wformat-nonliteral" but I prefer to use your patch instead.

adding ATTRIBUTE_PRINTF doesn't suppress warnings, it fixes them
-mike