On Fri, Mar 13, 2020 at 17:09:14 -0700, Paul Goyette wrote:
> On Sat, 14 Mar 2020, Valery Ushakov wrote:
>
> > On Fri, Mar 13, 2020 at 14:17:42 -0400, Christos Zoulas wrote:
> >
> > > Log Message:
> > > PR/55068: sc.dying: Fix printf formats:
> > [...]
> > > - 0x% -> %#
> >
> > This was not a part of the PR and is completely cosmetic (surely it
> > supports plain %x if it does support %#x). Why was this necessary?
> > (I know I would be quite miffed if someone made a change like that to
> > my code).
>
> Plain %x - no :(
>
> In order to enable sysctl-transport to userland, all the args need to
> be promoted to %jx, and the format strings need to ensure that they
> consume that size.
Random sample from the diff:
- printf("%s: expect 0xe6!! (0x%x)\n", device_xname(sc->sc_dev),
+ printf("%s: expect 0xe6!! (%#x)\n", device_xname(sc->sc_dev),
Actually, looking close I see diffs like
- DPRINTFN(MD_ROOT, "wValue=0x%04jx", value, 0, 0, 0);
+ DPRINTFN(MD_ROOT, "wValue=%#04jx", value, 0, 0, 0);
that are plain wrong as %#x counts the 0x prefix towards the field
width.
$ printf '0x%02x %#02x\n' 1 1
0x01 0x1
$ printf '0x%08x 0x%08x\n%#08x %#08x\n' 0 1 0 1
0x00000000 0x00000001
00000000 0x000001
So, as far as I can tell, these %# changes don't fix all the argument
size issues, break some output formatting and violate preference of
the original author. Did I miss something else?
-uwe
