Re: [PATCH] Added FTW.H Methods

[Corinna Vinschen <vinschen@redhat.com>]

On Aug 17 16:10, Eshan Dhawan via Newlib wrote:
> 
> > On 17-Aug-2020, at 3:12 PM, Corinna Vinschen <vinschen@redhat.com> wrote:
> > 
> > On Aug 13 08:11, Joel Sherrill wrote:
> >>> On Thu, Aug 13, 2020 at 3:17 AM Corinna Vinschen via Newlib <
> >>> newlib@sourceware.org> wrote:
> >>> 
> >>> On Aug 13 03:08, Eshan dhawan via Newlib wrote:
> >>>> Signed-off-by: Eshan dhawan <eshandhawan51@gmail.com>
> >>>> ---
> >>>> newlib/configure.host         |   2 +-
> >>>> newlib/libc/include/ftw.h     |  64 ++++++++++++++
> >>>> newlib/libc/posix/Makefile.am |   2 +-
> >>>> newlib/libc/posix/ftw.c       |  36 ++++++++
> >>>> newlib/libc/posix/nftw.c      | 154 ++++++++++++++++++++++++++++++++++
> >>>> 5 files changed, 256 insertions(+), 2 deletions(-)
> >>>> create mode 100644 newlib/libc/include/ftw.h
> >>>> create mode 100644 newlib/libc/posix/ftw.c
> >>>> create mode 100644 newlib/libc/posix/nftw.c
> >>> 
> >>> Why do we need _FTW_ENABLE_?
> >>> 
> >>> What do we need the foo64 variants for?  WHile newlib defines
> >>> _LARGEFILE64_SOURCE for __linux__ and __RDOS__, it's not used
> >>> anywhere else in newlib itself.
> >>> 
> >>> Also, there's a ftw/nftw implementation in Cygwin already.  Did you
> >>> check it for inclusion into newlib, or to drop the Cygwin version in
> >>> favor of a new newlib versions?  We could at least share a common
> >>> header, but then we will have to use the flag values already used in
> >>> the Cygwin implementation for backward compatibility.
> >>> 
> >> 
> >> We went straight to *BSD for implementations and didn't look at Cygwin.
> > 
> > Cygwin's ftw is OpenBSD.  Which BSD is this ftw taken from?
> It’s taken from MUSL Libc 

Ok.

> >> I've asked Eshan to replace the implementation with the Cygwin one.
> >> Hopefully that's easy and allows one more piece of shared software.
> > 
> > It's not *that* easy because OpenBSD's/Cygwin's ftw(3) uses fts(3) under
> > the hood.
> That’s the same reason We couldn’t use FTW from BSD 
> > 
> > While sharing code is good, your implementation is more compact, which
> > speaks for your code.
> > 
> > However, I'd like to ask for ftw.h compatibility at least, so perhaps
> > Cygwin can use this other ftw, too at one point.
> > 
> > But even then, there's still the _FTW_ENABLE_ and _LARGEFILE64_SOURCE
> > stuff I asked about.
> _FTW_ENABLE_ was there because the FTW implementation required pthread_setcancelstate ()

That puzzles me a bit.  Not setting _FTW_ENABLE_ disables ftw entirely,
just because it requires pthread_setcancelstate.  However, newlib
already supports a flag _STDIO_WITH_THREAD_CANCELLATION_SUPPORT to call
pthread_setcancelstate only on targets providing it, see, for instance,
newlib/libc/stdio/freopen.c.  ftw could simply do the same.


Thanks,
Corinna