Wenchao Xia <xiaw...@linux.vnet.ibm.com> writes:
> δΊ 2013-4-11 20:05, Markus Armbruster ει:
>> Wenchao Xia <xiaw...@linux.vnet.ibm.com> writes:
>>
>>>>>
>>>>> -void bdrv_image_info_dump(ImageInfo *info)
>>>>> +void bdrv_image_info_dump(GString *buf, ImageInfo *info)
>>>>> {
>>>>> char size_buf[128], dsize_buf[128];
>>>>> if (!info->has_actual_size) {
>>>>> @@ -370,43 +369,48 @@ void bdrv_image_info_dump(ImageInfo *info)
>>>>
>>>> I don't like this change, because it introduces buffering for no
>>>> discernible reason. Unless you can show me one, I'd like you to keep
>>>> printing directly.
>>>>
>>> HMP code later need to call this function, and then print buf to
>>> monitor console, which is the goal of this patch.
>>
>> Aha, the actual problem is "print either to stdout or to current
>> monitor", so that we can share code among qemu-img and monitor commands.
>> Such sharing is good, and the problem is real.
>>
>> bdrv_snapshot_dump() solves it this way: print to a fixed-size buffer,
>> which the caller either prints to stdout (qemu-img) or to the current
>> monitor (info snapshots).
>>
>> Your patch makes the buffer flexible. Improvement.
>>
>> But in my opinion, the detour through the buffer is silly. Why not
>> simply use a print function that does the right thing?
>>
>> We already have such a function for "print either to stderr or to
>> current monitor": error_printf(). Could easily add one for stdout.
>>
>> [...]
>>
> Indeed it is a better way, how about:
> void message_printf(const char *fmt, ...);
> I am not sure where should this function be placed,
> hmp.c?
I figure it needs to go in a file that's linked into the utilities, too.
I think these are in make variables util-obj-y, stub-obj-y.
I'd be tempted to put it into qemu-error.c. If you don't like that
because it's not about errors, consider renaming the file :)
