Re: Question about the use of _newlib_flockfile_* in newlib

[Corinna Vinschen <vinschen at redhat dot com>]

On Dec 18 10:35, Bin.Cheng wrote:
> On Tue, Dec 18, 2012 at 3:55 AM, Jeff Johnston <jjohnstn@redhat.com> wrote:
> > On 12/17/2012 05:53 AM, Corinna Vinschen wrote:
> >>
> >> On Dec 17 10:06, Bin.Cheng wrote:
> >>>
> >>> Hi,
> >>> I have two questions on _newlib_flockfile_*.
> >>> 1.
> >>> _fputc_r calls _putc_r to do the job, but both of them calls
> >>> _newlib_flockfile_*. Is it OK? Why if it is.
> >>
> >>
> >> I'm not quite sure, but I think the reason is that putc and, in analogy,
> >> _putc_r might be defined as macro.  Newlib doesn't do that by default
> >> for _putc_r, but target-specific code could.
> >>
> >>> 2. In most functions, the macro ORIENT() is guarded by
> >>> _newlib_flockfile_*, but it is not in _puts_r. Anything special or
> >>> just missed something?
> >>
> >>
> >> I had a look into the CVS annotation and it seems that this is not about
> >> ORIENT, but about the fact that the __sfvwrite_r call in _puts_r was
> >> never guarded by a flockfile/funlockfile (the predecessor of
> >> _newlib_flockfile_start/_newlib_flockfile_end).
> >>
> >> I don't know why, though.  Jeff?
> >>
> >
> > Don't know why this is the case.  Perhaps puts used to call fputs previously
> > which has the lock protection so the locks were never needed.  I can't see
> > any reason not to add them in the present case.
> 
> Hi Jeff,
> There is typo in your checkin for puts.c
> 
>   _newlib_flockfile_start (_stdout_r (ptr));
>   ORIENT (_stdout_r (ptr), -1);
>   result = (__sfvwrite_r (ptr, _stdout_r (ptr), &uio) ? EOF : '\n');
>   _newlib_flockfile_start (_stdout_r (ptr));
> 
> Thanks.

I checked in patch.  I added a local var fp to avoild calling
_stdout_r 4 times in a row.  Even if the compiler probably optimizes
that anyway, it's easier to read, imho.  Hope that's ok.


Corinna

-- 
Corinna Vinschen
Cygwin Project Co-Leader
Red Hat