Re: fatal: substr: length 0 is not >= 1

[arnold]

Hi.

Thanks for your note.

The lint option checking that the length parameter is zero assumes
(as in this case) that the result comes from a runtime computation and
that the user wants a real substring and not an empty string. In such a
case, a warning that zero came in is appropriate.

-Lfatal makes things a little worse than intended, in that the
program exits immediately.  Personally, I don't recommend running
with -Lfatal in production, but rather during development and testing.

If -Lfatal is the default, then you have to protect the substr()
call, as you've done.

Thanks,

Arnold

Roland Illig <address@hidden> wrote:

> Dear GNU Awk developers,
>
> In a real-world AWK program
> (https://github.com/NetBSD/pkgsrc/blob/8075fc03e305003a/mk/scripts/subst-identity.awk)
> I wrote the following code:
>
> # Tests whether a single "s,from,to," is an identity substitution.
> function is_identity_subst(
>       s,
>       len, i, sep, pat_from, pat_to, ch, subst
> ) {
>       len = length(s);
>       if (len < 6 || substr(s, 1, 1) != "s")
>               return 0;
>
>       sep = substr(s, 2, 1);
>       i = 3;
>       pat_to = "";
>       while (i < len && substr(s, i, 1) != sep) {
>               ch = identity_char(substr(s, i));
>               if (ch == "")
>                       break;
>               pat_to = pat_to substr(ch, 1, 1);
>               i += length(ch);
>       }
>
>       # The next 2 lines are only needed for GNU Awk 5.0.1
>       # in -Lfatal mode.
>       if (pat_to == "")
>               return 0;
>
>       pat_from = substr(s, 3, i - 3);
>
>       subst = "s" sep pat_from sep pat_to sep;
>       return s == subst || s == subst "g";
> }
>
>
> I ran this program with -Lfatal, and it presented me this "error":
>
> fatal: substr: length 0 is not >= 1
>
> I think it is ok to have length 0, which would always return the null
> string. In my mind, only negative lengths should trigger this error
> message. Was there a specific reason for making length 0 an error?
>
> I know that in the above code, I could have written the equivalent "i ==
> 3" instead. But that would have repeated the number 3, and I didn't want
> that. I thought was that the expression 'pat_to == ""' expressed my
> thoughts more clearly.
>
> Best,
> Roland
>