Re: Loaded footgun open_datasync on Windows

Kuntal Ghosh wrote:
[pg_test_fsync doesn't use pgwin32_open and hence doesn't respect O_DSYNC]
> On Wed, Jun 6, 2018 at 2:39 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> > On Wed, Jun 6, 2018 at 10:18 AM, Michael Paquier <michael(at)paquier(dot)xyz> wrote:
> > > On Wed, Jun 06, 2018 at 09:58:34AM +0530, Amit Kapila wrote:
> > > > One another alternative could be that we
> > > > define open as pgwin32_open (for WIN32) wherever we need it.
> > >
> > > Which is what basically happens on any *nix platform, are you foreseeing
> > > anything bad here?
> >
> > Nothing apparent, but I think we should try to find out why at the first
> > place this has been made backend specific.
>
> It seems the "#ifndef FRONTEND" restriction was added around
> pgwin32_open() for building libpq with Visual C++ or Borland C++. The
> restriction was added in commit 422d4819 to build libpq with VC++[1].
> Later, in commit fd7c3f67e0bc4, the support for Borland C++ was also
> added.
>
> So, I'm not sure whether removing that restriction will work for all
> front-end modules.

Thanks for the research, and sorry for my long silence.

If I remove the "#ifndef FRONTEND", building with MSVC fails for all
call sites that use the two-argument version of open(2).

If I use the three-argument version throughout, which should be no problem,
PostgreSQL builds just fine with MSVC.

How about checking what the buildfarm thinks about the attached?

Yours,
Laurenz Albe