[PATCH resend] webservices: Use sscanf to convert strings to doubles.

[Erich E. Hoover]

This removes the last "long double" instance impacted by commit
0e183cc3c0d3b6f89f79047cdd71c389afc75073.  Thanks go again to Piotr
for his help.

Next time I'll remember to wait for the VMs to update ;)

Best,
Erich

[Piotr Caban]

Hi Erich,

+    if (p[len-1] == 'e' || p[len-1] == 'E') return WS_E_INVALID_FORMAT; /* "1e" */
This is no longer needed. scanf is now returning error on such input.

Sorry I haven’t seen it earlier,
Piotr

[Hans Leidekker]

On Fri, 2020-01-03 at 17:08 -0700, Erich E. Hoover wrote:
> +    if (*p == '+' || *p == '-')
>      {
> -        tmp = val * 10 + p[i] - '0';
> -        if (val > MAX_UINT64 / 10 || tmp < val)
> +        if (len == 1 || (len == 2 && *(p+1) == '.'))
>          {
> -            for (; i < nb_digits; i++) exp++;
> -            break;
> +            ret = 0;
> +            return S_OK;
>          }

'ret' is a pointer.
> +        ret = 0;
> +        return S_OK;

Same here.
> +    for (i = 0; i < len; i++)
>      {
> -        if (exp & 1) exp_val *= exp_mul;
> -        exp_mul *= exp_mul;
> +        if (p[i] >= '0' && p[i] <= '9') continue;
> +        if (p[i] != 'e' && p[i] != 'E' && p[i] != '.' && p[i] != '+' && p[i] != '-')
> +            return WS_E_INVALID_FORMAT;
>      }

Is this part really necessary? I also get new failures with a non-mingw build:

reader.c:3729: Test failed: 0: got 803d0000
reader.c:3729: Test failed: 1: got 803d0000
reader.c:3729: Test failed: 2: got 803d0000
reader.c:3729: Test failed: 8: got 803d0000
reader.c:3729: Test failed: 9: got 803d0000
reader.c:3729: Test failed: 10: got 803d0000
...

Looks like scanf returns a different result for %n in this case. It probably
shouldn't block this patch though.

[Erich E. Hoover]

On Mon, Jan 6, 2020 at 3:52 AM Hans Leidekker <hans(a)codeweavers.com> wrote:
> On Fri, 2020-01-03 at 17:08 -0700, Erich E. Hoover wrote:
> > ...
> > +            ret = 0;
> > ...
>
> 'ret' is a pointer.

Ack, sorry about that - supposed to be setting the value.
> > ...
> > +    for (i = 0; i < len; i++)
> >      {
> > -        if (exp & 1) exp_val *= exp_mul;
> > -        exp_mul *= exp_mul;
> > +        if (p[i] >= '0' && p[i] <= '9') continue;
> > +        if (p[i] != 'e' && p[i] != 'E' && p[i] != '.' && p[i] != '+' && p[i] != '-')
> > +            return WS_E_INVALID_FORMAT;
> >      }
>
> Is this part really necessary?

Technically sscanf supports more than what the reader does, but it
does pass all the current tests without this check.  For example,
scanf supports the Fortran-style "d" exponent notation (for msvcr <
140).  If you would like I can fold the +/-/. weirdness into this,
which does have tests, possibly something like:
===
    for (i = 0; i < len; i++)
    {
        if (p[i] >= '0' && p[i] <= '9')
        {
            found_digit = TRUE;
            continue;
        }
        if (p[i] != 'e' && p[i] != 'E' && p[i] != '.' && p[i] != '+'
&& p[i] != '-')
            return WS_E_INVALID_FORMAT;
    }
    if (!found_digit)
    {
        *ret = 0;
        return S_OK;
    }
===
> I also get new failures with a non-mingw build:
> ...
> Looks like scanf returns a different result for %n in this case. It probably
> shouldn't block this patch though.

That's interesting, if you like I could move the length check into the
format checker so that it does not rely on scanf for that to work
properly.

Best,
Erich

[Erich E. Hoover]

On Mon, Jan 6, 2020 at 9:35 AM Erich E. Hoover <erich.e.hoover(a)gmail.com> wrote:
> ...
> That's interesting, if you like I could move the length check into the
> format checker so that it does not rely on scanf for that to work
> properly.

For example, the attached version.

Best,
Erich