Re: [Qemu-devel] [PATCH v5 13/20] hw/acpi/aml-build: Add ToUUID macro

[Shannon Zhao <shannon.zhao@linaro.org>]

On 2015/5/4 17:22, Igor Mammedov wrote:
> On Wed, 29 Apr 2015 21:41:39 +0800
> Shannon Zhao <shannon.zhao@linaro.org> wrote:
> 
>>
>>
>> On 2015/4/28 17:48, Shannon Zhao wrote:
>>> On 2015/4/28 17:35, Igor Mammedov wrote:
>>>> On Tue, 28 Apr 2015 16:52:19 +0800
>>>> Shannon Zhao <zhaoshenglong@huawei.com> wrote:
>>>>
>>>>> On 2015/4/28 16:15, Igor Mammedov wrote:
>>>>>>>> btw:
>>>>>>>>>> whole thing might be simpler if an intermediate conversion is avoided,
>>>>>>>>>> just pack buffer as in spec byte by byte:
>>>>>>>>>>
>>>>>>>>>> /* format: aabbccdd-eeff-gghh-iijj-kkllmmnnoopp */
>>>>>>>>>> assert(strlen(uuid) == ...);
>>>>>>>>>> build_append_byte(var->buf, HEX2BYTE(uuid[3]); /* dd */
>>>>>>>>
>>>>>>>> use build_append_byte(var->buf, HEX2BYTE(uuid + 7); ?
>>>>>>>>
>>>>>>>>>> build_append_byte(var->buf, HEX2BYTE(uuid[2]); /* cc */
>>>>>>>>
>>>>>>>> use build_append_byte(var->buf, HEX2BYTE(uuid + 5); ?
>>>>>> if you mean hyphens /-/ then they are not encoded,
>>>>>> but you surely can add checks for them to make sure that
>>>>>> UUID format is as expected.
>>>>>>
>>>>>
>>>>> I mean uuid[3] points to b not dd. Maybe use following way:
>>>>>
>>>>> static uint8_t Hex2Byte(char *src)
>>>> or even better:
>>>> Hex2Byte(char *src, byte_idx)
>>>>    and do pointer arithmetic inside
>>>>
>>>> [...]
>>>>> build_append_byte(var->buf, Hex2Byte(uuid + (3 * 2))); /* dd */
>>>> build_append_byte(var->buf, Hex2Byte(uuid, 3)); /* dd - at offset 00 */
>>>> build_append_byte(var->buf, Hex2Byte(uuid, 2)); /* cc - at offset 01 */
>>>> ...
>>>>
>>> Yes, it's better to first four bytes. But there are hyphens /-/, for
>>> offset 04, 05 and etc it doesn't work. We can't use following expression:
>>>
>>> build_append_byte(var->buf, Hex2Byte(uuid, 5)); /* ff - at offset 04 */
>>> build_append_byte(var->buf, Hex2Byte(uuid, 4)); /* ee - at offset 05 */
>>> ...
>>>
>>>
>>
>> So about the implementation of this macro, I think I'd like to use 
>> following approach. This lets Hex2Byte do what it should only do and 
>> still has a clear look of UUID. What do you think about?
>>
>> static uint8_t Hex2Byte(char *src)
>> {
>>      int hi = Hex2Digit(*src++);
>>      int lo = Hex2Digit(*src);
>>
>>      if ((hi < 0) || (lo < 0))
>>          return -1;
> just make Hex2Digit() assert on wrong input.
> 

Ok.

>>
>>      return (hi << 4) | lo;
>> }
>>
>> g_assert((strlen(uuid) == 36) && (uuid[8] == '-') && (uuid[13] == '-')
>>            && (uuid[18] == '-') && (uuid[23] == '-'));
>>
>> build_append_byte(var->buf, Hex2Byte(uuid + (3 * 2))); /* dd */
>                                                ^^^^^
> I'd make it one number, instead of forcing reader to do math
> every time he/she looks at this code.
> 

Ok, will do this. And about other patches in this series could you offer
your comments?

Thanks,
Shannon

> 
>> build_append_byte(var->buf, Hex2Byte(uuid + (2 * 2))); /* cc */
>> build_append_byte(var->buf, Hex2Byte(uuid + (1 * 2))); /* bb */
>> build_append_byte(var->buf, Hex2Byte(uuid + (0 * 2))); /* aa */
>>
>> build_append_byte(var->buf, Hex2Byte(uuid + (5 * 2 + 1))); /* ff */
>> build_append_byte(var->buf, Hex2Byte(uuid + (4 * 2 + 1))); /* ee */
>>
>> build_append_byte(var->buf, Hex2Byte(uuid + (7 * 2 + 2))); /* hh */
>> build_append_byte(var->buf, Hex2Byte(uuid + (6 * 2 + 2))); /* gg */
>>
>> build_append_byte(var->buf, Hex2Byte(uuid + (8 * 2 + 3))); /* ii */
>> build_append_byte(var->buf, Hex2Byte(uuid + (9 * 2 + 3))); /* jj */
>>
>> build_append_byte(var->buf, Hex2Byte(uuid + (10 * 2 + 4))); /* kk */
>> build_append_byte(var->buf, Hex2Byte(uuid + (11 * 2 + 4))); /* ll */
>> build_append_byte(var->buf, Hex2Byte(uuid + (12 * 2 + 4))); /* mm */
>> build_append_byte(var->buf, Hex2Byte(uuid + (13 * 2 + 4))); /* nn */
>> build_append_byte(var->buf, Hex2Byte(uuid + (14 * 2 + 4))); /* oo */
>> build_append_byte(var->buf, Hex2Byte(uuid + (15 * 2 + 4))); /* pp */
>