On 25/10/2016 13:06, Halil Pasic wrote:
>
>
> On 10/24/2016 06:31 PM, Paolo Bonzini wrote:
>> Implement error_vprintf to send the output of error_report to
>> the test log. This silences test-vmstate.
>>
>> Signed-off-by: Paolo Bonzini <[email protected]>
> [..]
>> void error_vprintf(const char *fmt, va_list ap)
>> {
>> - vfprintf(stderr, fmt, ap);
>> + if (g_test_initialized() && !g_test_subprocess()) {
>
> I like the approach. What I do not like is:
> * we still generate noise only less visible
It's not noise if it ends up exactly in the right place (the test logs). :)
Paolo
> * it's indiscriminate (silences all error reports of this type)
> * can't be incorporated into the test logic
>
> IMHO these are no major problems, and what I'm going to suggest
> might be over-engineering, but I will ask nevertheless.
>
> Have you considered a thread local function pointer to
> defaulting to something sane (eg. to this test log stuff
> or to a wrapped vprintf)?
>
> This would invalidate all the points above, and make the compat
> macros unnecessary -- which have bitten me on my ancient RHEL
> (because glib is too old). I have tried out the idea, so
> if you want I can send you a patch on top of this.
>
> Halil
>
>> + char *msg = g_strdup_vprintf(fmt, ap);
>> + g_test_message("%s", msg);
>> + g_free(msg);
>> + } else {
>> + vfprintf(stderr, fmt, ap);
>> + }
>> }
>>
>> void error_vprintf_unless_qmp(const char *fmt, va_list ap)
>>
>
