Re: Coverity Scan: Analysis completed for RTEMS-Newlib

On Nov 11 00:59, Takashi Yano wrote:
> On Sat, 11 Nov 2023 00:55:15 +0900
> Takashi Yano wrote:
> > On Fri, 10 Nov 2023 16:31:44 +0100
> > Corinna Vinschen wrote:
> > > On Nov 10 08:50, Joel Sherrill wrote:
> > > > Hmmmm.. an email just before the one I forwarded shows 6 new defects were
> > > > added in the last commits. They appear to be the same issue I just
> > > > forwarded but in different scanf variants.
> > > > 
> > > > CID 423229 (#1 of 1): Uninitialized scalar variable (UNINIT)2.
> > > > uninit_use_in_call: Using uninitialized value f._flags2 when calling
> > > > __ssvfiscanf_r. [show details
> > > > <https://scan3.scan.coverity.com/eventId=13202494-1&modelId=13202494-0&fileInstanceId=104130545&filePath=%2Fhome%2Fjoel%2Frtems-cron-coverity%2Fsourceware-mirror-newlib-cygwin%2Fnewlib%2Flibc%2Fstdio%2Fvfscanf.c&fileStart=400&fileEnd=1980>
> > > > ]
> > > 
> > > Looks like a false positive.
> > > 
> > > The ORIENT macro will set or reset the value of the _flags2 __SWID bit
> > > if the _flags __SORD bit isn't set.  It never is set at the start, so
> > > the _flags2 __SWID bit is always set.  And only then, the ORIENT macro
> > > will check the value.
> > 
> > Perhaps, this happens because other bits of _flags2 than __SWID is not
> > initialized.
> > 
> > Which is better solution do you think?
> > (1) Modify ORIENT macro so that it returns (ori > 0) ? 1 : -1.
> > (2) Initialize f._flags2 = 0 in sscanf() family.
> 
> Ah, this problem will also occur for sprintf() family. So it seems that
> (1) is easier becase f._file = -1 is set at 41 places.

I wonder, though, if setting f._flags2 = 0 isn't all around the better
solution.  That would mean that ORIENT itself doesn't have to become
more complicated, having to check (ori > 0) even in places, where this
isn't a problem.


Corinna