Re: [PATCH] parser: Fix VSLENGTH parsing with trailing garbage

[Denys Vlasenko <vda.linux@xxxxxxxxxxxxxx>]

On Tue, Jun 22, 2021 at 2:19 AM Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote:
> On Mon, Jun 21, 2021 at 04:21:40PM +0200, Denys Vlasenko wrote:
> > > -                       } while (!subtype && is_digit(c));
> > > +                       } while ((subtype <= 0 || subtype >= VSLENGTH) &&
> > > +                                is_digit(c));
> >
> > ... you use (subtype == 0 || subtype == VSLENGTH) here.
> > Also, (subtype == 0 || subtype == VSLENGTH) is less confusing:
> > it says "loop if ${VAR} or ${#VAR} syntax", whereas <= >=
> > are a bit misleading.
>
> Yes it looks a bit confusing, but it turns into a single branch
> instead of two.

Yes, I know that. Compiler turns it into "(unsigned)(x-1) >= VSLENGTH-1"
expression.

But is it worth the obfuscation? Especially that it also has another
downside (it requires an additional free CPU register to hold (x-1)
result, which can force compiler to spill other values to stack).