Re: [PATCH v2 1/1] utils: Use fixed-point arithmetic in qemu_strtosz

[Eric Blake <eblake@redhat.com>]

On 3/17/21 6:16 AM, Eric Blake wrote:
> On 3/17/21 2:09 AM, Thomas Huth wrote:
>> On 15/03/2021 16.58, Richard Henderson wrote:
>>> Once we've parsed the fractional value, extract it into an integral
>>> 64-bit fraction.  Perform the scaling with integer arithemetic, and
>>> simplify the overflow detection.
>>
>> I've put this patch in my local branch, but I'm still getting a failure
>> in the cutils test, this time in the Cirrus-CI with the MinGW build:
>>
>>  https://cirrus-ci.com/task/5413753530351616?command=test#L543
>>
>> Is it related or is this a different bug?
> 
> ERROR test-cutils - Bail out!
> ERROR:../tests/unit/test-cutils.c:2233:test_qemu_strtosz_trailing:
> assertion failed (res == 0): (1024 == 0)
> 
> That's testing behavior on:
> 
>     str = "0x";
>     err = qemu_strtosz(str, &endptr, &res);
> 
> which should parse as "0" with a suffix of 'x'.  It is an independent
> issue (unrelated to the rounding issues fixed in rth's patch), and
> rather appears to be a bug in mingw's libc for strtoull although I have
> not actually set up an environment to test that assumption yet.

Confirmed:
$ cat foo.c
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>

int main(void)
{
    unsigned long u;
    const char *str = "0x";
    char *end;
    errno = 0;
    u = strtoul(str, &end, 10);
    printf("%lu %s %d\n", u, end, errno);
    errno = 0;
    u = strtoul(str, &end, 16);
    printf("%lu %s %d\n", u, end, errno);

    str = "0xq";
    errno = 0;
    u = strtoul(str, &end, 10);
    printf("%lu %s %d\n", u, end, errno);
    errno = 0;
    u = strtoul(str, &end, 16);
    printf("%lu %s %d\n", u, end, errno);
    return 0;
}
$ gcc -o foo-linux -Wall foo.c
$ x86_64-w64-mingw32-gcc -o foo-mingw -Wall foo.c
$ ./foo-linux
0 x 0
0 x 0
0 xq 0
0 xq 0
$ wine ./foo-mingw 2>/dev/null
0 x 0
0 0x 0
0 xq 0
0 0xq 0

Mingw has a bug (and therefore so do all our qemu_strto* functions) when
parsing "0xgarbage" with a base of 0 or 16, in that it fails to advance
past the leading '0' (which is a valid parse).  Patch coming up.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org