From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <axboe@kernel.dk>
Message-ID: <55918E78.2050706@kernel.dk>
Date: Mon, 29 Jun 2015 12:29:12 -0600
From: Jens Axboe <axboe@kernel.dk>
MIME-Version: 1.0
Subject: Re: competime_assert failure without -O3 optimization flag expected
 ?
References: <CAB-428=biE+PMVOA8qg_VcH9bf0ZVZO1dimXVh3KC3Z7j=QrmQ@mail.gmail.com> <5591687D.6040803@kernel.dk> <CAB-428=EKjcxK6J9OvKU1nM34Q998fN=LFQKgNgofeMb_cuY1g@mail.gmail.com> <55916A66.5080303@kernel.dk> <55916C87.1030408@kernel.dk> <CAB-428=ToOGYb=5r6VhORWiFqXoHk6N8tMX8HbUr2_Cctf9oWw@mail.gmail.com> <55918558.8090006@kernel.dk> <CAB-428=30M7HXTfL4P3KpyjAG3dqMA9vVirc8iyjbO-8eAgDrQ@mail.gmail.com>
In-Reply-To: <CAB-428=30M7HXTfL4P3KpyjAG3dqMA9vVirc8iyjbO-8eAgDrQ@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 7bit
To: Alireza Haghdoost <haghdoost@gmail.com>
Cc: "fio@vger.kernel.org" <fio@vger.kernel.org>
List-ID: <fio@vger.kernel.org>

On 06/29/2015 12:25 PM, Alireza Haghdoost wrote:
> On Mon, Jun 29, 2015 at 12:50 PM, Jens Axboe <axboe@kernel.dk> wrote:
>> On 06/29/2015 11:42 AM, Alireza Haghdoost wrote:
>>>
>>> Jens,
>>>
>>> Sounds like the compiletime_assert() method was not function and is
>>> relying on the optimizer performing dead code elimination to remove
>>> the call to prefix ## suffix
>>>
>>> How about this solution: I guess we have to work around it to make it
>>> more portable since it works with C11.
>>>
>>> diff --git a/fio-arh/compiler/compiler.h b/fio-arh/compiler/compiler.h
>>> index 40e857c..7c9ba57 100644
>>> --- a/fio-arh/compiler/compiler.h
>>> +++ b/fio-arh/compiler/compiler.h
>>> @@ -33,26 +33,6 @@
>>>     1; \
>>>    })
>>>
>>> -#ifndef __compiletime_error
>>> -#define __compiletime_error(message)
>>> -#endif
>>> -#ifndef __compiletime_error_fallback
>>> -#define __compiletime_error_fallback(condition) do { } while (0)
>>> -#endif
>>> -
>>> -#define __compiletime_assert(condition, msg, prefix, suffix) \
>>> - do { \
>>> - int __cond = !(condition); \
>>> - extern void prefix ## suffix(void) __compiletime_error(msg); \
>>> - if (__cond) \
>>> - prefix ## suffix(); \
>>> - __compiletime_error_fallback(__cond); \
>>> - } while (0)
>>> -
>>> -#define _compiletime_assert(condition, msg, prefix, suffix) \
>>> - __compiletime_assert(condition, msg, prefix, suffix)
>>> -
>>> -#define compiletime_assert(condition, msg) \
>>> - _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
>>> +#define compiletime_assert(condition, msg) _Static_assert(condition, msg)
>>>
>>>    #endif
>>
>>
>> But now it's more compiler dependent, which is worse than before. At least
>> it only broke if people fiddled with the optimizations before, otherwise it
>> was fine.
>>
>> Add a configure test for this, use _Static_assert() if it's available, the
>> old method if not.
>
> Here you are:
>
> diff --git a/fio-arh/compiler/compiler.h b/fio-arh/compiler/compiler.h
> index 40e857c..93fdc56 100644
> --- a/fio-arh/compiler/compiler.h
> +++ b/fio-arh/compiler/compiler.h
> @@ -1,5 +1,6 @@
>   #ifndef FIO_COMPILER_H
>   #define FIO_COMPILER_H
> +#include <assert.h>
>
>   #if __GNUC__ >= 4
>   #include "compiler-gcc4.h"
> @@ -33,6 +34,12 @@
>    1; \
>   })
>
> +
> +#if (__STDC_VERSION__ >= 201112L)
> +#define compiletime_assert(condition, msg) _Static_assert(condition, msg)
> +#else
> +
> +
>   #ifndef __compiletime_error
>   #define __compiletime_error(message)
>   #endif
> @@ -55,4 +62,7 @@
>   #define compiletime_assert(condition, msg) \
>    _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
>
> +
> +#endif
> +
>   #endif
>

This is not going to work for earlier compilers, in fact it breaks on 
even gcc 4.9 here. As I said, this needs to be a configure test. That is 
a lot more reliable than this sort of version checking.

-- 
Jens Axboe