Re: [Qemu-devel] [PATCH v1 1/9] cutils: add qemu_strtod() and qemu_strtod_finite()

[Eric Blake]

On 11/15/18 8:04 AM, David Hildenbrand wrote:
> Let's provide a wrapper for strtod().
>
> Signed-off-by: David Hildenbrand <address@hidden>
> ---
>   include/qemu/cutils.h |  2 ++
>   util/cutils.c         | 38 ++++++++++++++++++++++++++++++++++++++
>   2 files changed, 40 insertions(+)


> +
> +/**
> + * Convert string @nptr to a finite double.
> + *
> + * Works like qemu_strtoul(), except it stores +/-HUGE_VAL on
> + * overflow/underflow. "NaN" or "inf" are rejcted with -EINVAL.

s/rejcted/rejected/

> + */
> +int qemu_strtod_finite(const char *nptr, const char **endptr, double *result)
> +{
> +    int ret = qemu_strtod(nptr, endptr, result);

On overflow, result is set to HUGE_VAL (aka "inf") with ret set to 
-ERANGE.  (The C standard uses HUGE_VAL rather than directly requiring 
infinity on overflow, in order to cater to museum platforms where the 
largest representable double is still finite; but no one develops qemu 
on a non-IEEE machine these days so we know that HUGE_VAL == INF).

> +
> +    if (!ret && !isfinite(*result)) {
> +        return -EINVAL;
> +    }

This check means that overflow ("1e9999") fails with -ERANGE, while 
actual infinity ("inf") fails with -EINVAL, letting the user distinguish 
between the two.  Still, I wonder if assigning a non-finite value into 
result on -ERANGE is the wisest course of action.  We'll just have to 
see in the next patches that use this.

With the typo fix,

Reviewed-by: Eric Blake <address@hidden>

--
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org