On Wed, Mar 22, 2017 at 05:53:00PM +0100, Max Reitz wrote:
> On 22.03.2017 17:44, Stefan Hajnoczi wrote:
> > On Mon, Mar 20, 2017 at 04:11:24PM +0100, Max Reitz wrote:
> >> On 20.03.2017 12:00, Stefan Hajnoczi wrote:
> >>> On Mon, Mar 13, 2017 at 10:40:36PM +0100, Max Reitz wrote:
> >>>> +static int raw_regular_truncate(int fd, BlockDriverState *bs, int64_t offset,
> >>>
> >>> The presence of both a file descriptor and a BlockDriverState (actually
> >>> it must be a BDRVRawState) arguments is unusual.  It seems bs is needed
> >>> for bdrv_getlength().
> >>>
> >>> How about using fstat(fd, &st) and then dropping bs and create?
> >>
> >> Well, I could do that, but bdrv_getlength() is a bit simpler and I don't
> >> see much of an issue with specifying both an fd and a bs.
> > 
> > Arguments that provide overlapping information make the function harder
> > to understand and use correctly (there are combinations of these
> > arguments that are invalid or don't make sense).  Saving a few lines of
> > code in the function implementation is a poor trade-off IMO.
> 
> My brain likes to agree but for some reason my heart really prefers
> block layer functions (where I know what can go wrong) over POSIX
> functions (where I always have the feeling of maybe not having though of
> everything).
> 
> I guess I'll have to silence my heart for a bit, then.

In matters of the heart I cannot give advice on this mailing list,
sorry.

Stefan