[dpdk-dev] [EXT] Re: [PATCH] tracepoint: fix compilation with C++

[Sunil Kumar Kori]

Answers inline.

Regards
Sunil Kumar Kori

>-----Original Message-----
>From: Pawel Wodkowski <pawelwod at gmail.com>
>Sent: Friday, August 7, 2020 12:14 AM
>To: Sunil Kumar Kori <skori at marvell.com>; Jerin Jacob Kollanukkaran
><jerinj at marvell.com>
>Cc: dev <dev at dpdk.org>; David Marchand <david.marchand at redhat.com>
>Subject: Re: [EXT] Re: [dpdk-dev] [PATCH] tracepoint: fix compilation with C++
>
>On 06.08.2020 08:44, Sunil Kumar Kori wrote:
>> Hello Pawel,
>>
>> See answers inline.
>>
>> Regards
>> Sunil Kumar Kori
>>
>>> -----Original Message-----
>>> From: David Marchand <david.marchand at redhat.com>
>>> Sent: Wednesday, August 5, 2020 1:30 PM
>>> To: Pawel Wodkowski <pawelwod at gmail.com>
>>> Cc: dev <dev at dpdk.org>; Jerin Jacob Kollanukkaran
>>> <jerinj at marvell.com>; Sunil Kumar Kori <skori at marvell.com>
>>> Subject: [EXT] Re: [dpdk-dev] [PATCH] tracepoint: fix compilation
>>> with C++
>>>
>>> External Email
>>>
>>> ---------------------------------------------------------------------
>>> -
>>> Hello Pawel,
>>>
>>> Thanks for contributing to DPDK.
>>>
>>>
>>> On Tue, Aug 4, 2020 at 7:52 PM Pawel Wodkowski
><pawelwod at gmail.com>
>>> wrote:
>>>> trace_mem is declared as 'void *' which triggers following error:
>>>> '...invalid conversion from ‘void*’ to ‘__rte_trace_header*’
>>>> [-fpermissive]...'
>>>>
>>>> Fix this by changing void to struct __rte_trace_header
>> trace_mem is intentionally kept as void * so that it can not be accessed by
>application directly as it part of global header.
>But this structure is well defined in this file anyway. It can be casted to 'struct
>__rte_trace_header *' and used.
>Isn't the double underscore prefix good enough warning that it is internal
>datatype?
>
>But anyway, I'm not here to discuss the architecture of DPDK so lets go to the
>next point.
>
>> If I understood the problem correctly, it is because of using
>> trace_mem without typecasting and GCC does not report it as error due to
>implicit typecast and G++ reports it as error as it does not do implicit
>typecasting.
>Actually it is because in C++ implicit cast from void pointer is not allowed. So
>when it is used like this
>
>      struct __rte_trace_header *trace = RTE_PER_LCORE(trace_mem);
>
>it triggers this type of error:
>
>include/rte_trace_point.h: In function ‘void*
>__rte_trace_mem_get(uint64_t)’:
>include/rte_per_lcore.h:44:46: error: invalid conversion from ‘void*’ to
>‘__rte_trace_header*’ [-fpermissive]
>  #define RTE_PER_LCORE(name) (per_lcore_##name)
>                                         ^
>nclude/rte_trace_point.h:303:37: note: in expansion of macro
>‘RTE_PER_LCORE’
>   struct __rte_trace_header *trace = RTE_PER_LCORE(trace_mem);
>                                                           ^
>
>One can add '-fpermisive' to allow this type of casting but it is only a
>workaround in C++ code. As you mentioned, other solution is typecast to
>__rte_trace_header but this is not needed in C and it have "__" prefix this why
>I decided to change 'void *' into 'struct __rte_trace_header *'.
>

But changing 'void *' to 'struct __rte_trace_header *' will change the interpretation of trace_mem from user perspective which is against the design.
I would request you to go for the proposed solution. Although  there is no need of this for GCC but if there is no harm. 

Also consider a use case where system is designed like  'structure is exposed to application as opaque pointer(void *) and typecasted by the underlying implementation before using it'.
Then changing opaque pointer to structure type is not expected.

>>
>>
>> If this is the case then, I think it is better to typecast the trace_mem where
>ever it is being used. Anyways that will be safe for both GCC and G++.
>> @Jerin Jacob Kollanukkaran Please suggest. If you have some thing mind.
>
>I'm fine any solution that make this code compile with C++. Please let me
>know what is the decision then I can make V2.
>
>Paweł
>>
>>
>> [snippet]
>> --
>>