Re: [PATCH v1 09/48] perf tests: Silence -Wshorten-64-to-32 warnings

[Namhyung Kim]

On Wed, Apr 02, 2025 at 09:53:57AM -0700, Ian Rogers wrote:
> On Wed, Apr 2, 2025 at 9:38 AM Leo Yan <leo.yan@xxxxxxx> wrote:
> >
> > Hi Ian,
> >
> > On Wed, Apr 02, 2025 at 08:42:58AM -0700, Ian Rogers wrote:
> >
> > [...]
> >
> > > On Wed, Apr 2, 2025 at 7:35 AM Leo Yan <leo.yan@xxxxxxx> wrote:
> > > >
> > > > On Tue, Apr 01, 2025 at 11:23:07AM -0700, Ian Rogers wrote:
> > > >
> > > > [...]
> > > >
> > > > > @@ -478,13 +478,14 @@ static int __cmd_test(struct test_suite **suites, int argc, const char *argv[],
> > > > >       int err = 0;
> > > > >
> > > > >       for (struct test_suite **t = suites; *t; t++) {
> > > > > -             int i, len = strlen(test_description(*t, -1));
> > > > > +             int i;
> > > > > +             int len = (int)strlen(test_description(*t, -1));
> > > >
> > > > Thanks for huge polish.
> > > >
> > > > Just a concern from me.  Throughout this patch, the methodology is not
> > > > consistent.  Some changes update variable types which is fine for me.
> > > >
> > > > But the case above it simply cast size_t to int.  Should we update the
> > > > variable type as 'size_t' at here?
> > >
> > > Thanks Leo, I played around with it, but don't mind if someone wants
> > > to do it a different way. I was trying to make the changes minimal.
> > > The problem typically with size_t is we then use the value, for
> > > example, as a printf size modifier and need to introduce lots of casts
> > > back to being an int.
> >
> > This conclusion is not quite right, see:
> > https://stackoverflow.com/questions/2524611/how-can-one-print-a-size-t-variable-portably-using-the-printf-family
> >
> > > When this isn't too great I've done it, but in
> > > this case I think keeping the int, the lack of casts but a cast here
> > > to capture that we expect test descriptions to fit in the size of an
> > > int is the least worst option.
> >
> > I would say in another way.  After we enabled a compiler warning
> > option, this will give us a chance to improve code by dismissing
> > the warnings (and avoid potential bugs).
> >
> > If we use casts just to silence warnings, we also lose the opportunity
> > to improve code quality.  The changes in this series that fix type
> > mismatches are good, but I think the use of casts is not particularly
> > helpful - we're simply switching from implicit compiler casts to
> > explicit ones.
> 
> Right, but I think changing function parameters, return types would
> turn into an epic refactor and the patch series is already unwieldy.

Yep, it's important to reduce the number and the size of patches from
the reviewer's pespective. :)


> With the casts we can see the behavior is deliberate but that's not to
> say it couldn't be better. With the warnings gone it allows
> -Wshorten-64-to-32 to find the truly errant 64 to 32 bit implicit

Do we want to enable this warning?  Is it only available on clang?


> casts. Anyway, I don't have time to do such a larger refactor so
> somebody else is going to need to pick that up. I did refactor the
> cases where I thought it mattered more, but as you say that does lead
> to a feeling of inconsistency in the series.

I'm curious there are any actual errorenous cases other than the return
type of comparisons from Leo.

Thnaks,
Namhyung