Re: [PATCH 2/7] Cygwin: Enable backward binary compatibility

On May 16 07:31, Sebastian Huber wrote:
> On 13/05/2022 22:15, Corinna Vinschen wrote:
> > On May 13 15:46, Sebastian Huber wrote:
> > > ---
> > >   winsup/cygwin/include/cygwin/config.h | 1 +
> > >   1 file changed, 1 insertion(+)
> > > 
> > > diff --git a/winsup/cygwin/include/cygwin/config.h b/winsup/cygwin/include/cygwin/config.h
> > > index 71a216fbd..2d410a2e7 100644
> > > --- a/winsup/cygwin/include/cygwin/config.h
> > > +++ b/winsup/cygwin/include/cygwin/config.h
> > > @@ -80,6 +80,7 @@ extern inline struct _reent *__getreent (void)
> > >   #define __TM_GMTOFF tm_gmtoff
> > >   #define __TM_ZONE   tm_zone
> > >   #define _USE_LONG_TIME_T 1
> > > +#define _NEWLIB_BACKWARD_BINARY_COMPAT 1
> > Why?
> > 
> > AFAICS we don't expose _REENT to user space.  In theory (knocking
> > on wood here), we shouldn't need _REENT binary compat.
> 
> Ok good, I always thought the unused members in struct _reent were mainly
> for Cygwin.

Sorry, my bad. I wasn't thinking straight last week, apparently.

While Cygwin doesn't directly expose the functions taking a reent ptr as
parameter, we have to expose some of them indirectly, because they are
used in the stdio macros and inline functions, for instance, __swbuf_r.

So, yeah, Cygwin actually needs _reent backward compat here.

> Maybe the option should be named --enable-newlib-reent-binary-compat and
> restricted to struct _reent.

That would be great.  Does the above require a v3 of your patch set?


Thanks,
Corinna