[PATCH] Do not expand macros to 'defined'

[PATCH] Do not expand macros to 'defined'

[Павел Крюков]

Expanding a macro which contains 'defined' PP keyword is UB.

sim/common/Changelog:
2019-01-16  Pavel I. Kryukov  <kryukov@frtk.ru>

        * sim-arange.c: eliminate DEFINE_NON_INLINE_P
---
 sim/common/sim-arange.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/sim/common/sim-arange.c b/sim/common/sim-arange.c
index 6373b742ce8..b3488ab564a 100644
--- a/sim/common/sim-arange.c
+++ b/sim/common/sim-arange.c
@@ -32,10 +32,7 @@ along with this program.  If not, see
<http://www.gnu.org/licenses/>.  */
 #include <string.h>
 #endif

-#define DEFINE_INLINE_P (! defined (SIM_ARANGE_C_INCLUDED))
-#define DEFINE_NON_INLINE_P defined (SIM_ARANGE_C_INCLUDED)
-
-#if DEFINE_NON_INLINE_P
+#ifdef SIM_ARANGE_C_INCLUDED

 /* Insert a range.  */

@@ -280,9 +277,7 @@ sim_addr_range_delete (ADDR_RANGE *ar,
address_word start, address_word end)
   build_search_tree (ar);
 }

-#endif /* DEFINE_NON_INLINE_P */
-
-#if DEFINE_INLINE_P
+#else /* SIM_ARANGE_C_INCLUDED */

 SIM_ARANGE_INLINE int
 sim_addr_range_hit_p (ADDR_RANGE *ar, address_word addr)
@@ -301,4 +296,4 @@ sim_addr_range_hit_p (ADDR_RANGE *ar, address_word addr)
   return 0;
 }

-#endif /* DEFINE_INLINE_P */
+#endif /* SIM_ARANGE_C_INCLUDED */

Re: [PATCH] Do not expand macros to 'defined'

[Simon Marchi]

On 2019-01-16 01:34, Павел Крюков wrote:
> Expanding a macro which contains 'defined' PP keyword is UB.
> 
> sim/common/Changelog:
> 2019-01-16  Pavel I. Kryukov  <kryukov@frtk.ru>
> 
>         * sim-arange.c: eliminate DEFINE_NON_INLINE_P
> ---
>  sim/common/sim-arange.c | 11 +++--------
>  1 file changed, 3 insertions(+), 8 deletions(-)
> 
> diff --git a/sim/common/sim-arange.c b/sim/common/sim-arange.c
> index 6373b742ce8..b3488ab564a 100644
> --- a/sim/common/sim-arange.c
> +++ b/sim/common/sim-arange.c
> @@ -32,10 +32,7 @@ along with this program.  If not, see
> <http://www.gnu.org/licenses/>.  */
>  #include <string.h>
>  #endif
> 
> -#define DEFINE_INLINE_P (! defined (SIM_ARANGE_C_INCLUDED))
> -#define DEFINE_NON_INLINE_P defined (SIM_ARANGE_C_INCLUDED)
> -
> -#if DEFINE_NON_INLINE_P
> +#ifdef SIM_ARANGE_C_INCLUDED
> 
>  /* Insert a range.  */
> 
> @@ -280,9 +277,7 @@ sim_addr_range_delete (ADDR_RANGE *ar,
> address_word start, address_word end)
>    build_search_tree (ar);
>  }
> 
> -#endif /* DEFINE_NON_INLINE_P */
> -
> -#if DEFINE_INLINE_P
> +#else /* SIM_ARANGE_C_INCLUDED */
> 
>  SIM_ARANGE_INLINE int
>  sim_addr_range_hit_p (ADDR_RANGE *ar, address_word addr)
> @@ -301,4 +296,4 @@ sim_addr_range_hit_p (ADDR_RANGE *ar, address_word 
> addr)
>    return 0;
>  }
> 
> -#endif /* DEFINE_INLINE_P */
> +#endif /* SIM_ARANGE_C_INCLUDED */

The patch LGTM, I agree these macros are not really needed.  But I am 
confused, does this actually fix your c++ build?  You mentioned you were 
using g++ (and therefore cpp), which supposedly can handle this fine:

https://gcc.gnu.org/onlinedocs/cpp/Defined.html

If so, what did the error look like?

Simon

Re: [PATCH] Do not expand macros to 'defined'

[Pavel Kryukov]

> which supposedly can handle this fine

Right, but not with '-Wall -Wextra -Werror' flags.
These pedantic options help us (a little) keeping C++ project portable
between compilers, although we do not intent to have this property on GDB integration.

--
Pavel

> 16 янв. 2019 г., в 23:02, Simon Marchi <simon.marchi@polymtl.ca> написал(а):
> 
>> On 2019-01-16 01:34, Павел Крюков wrote:
>> Expanding a macro which contains 'defined' PP keyword is UB.
>> sim/common/Changelog:
>> 2019-01-16  Pavel I. Kryukov  <kryukov@frtk.ru>
>>        * sim-arange.c: eliminate DEFINE_NON_INLINE_P
>> ---
>> sim/common/sim-arange.c | 11 +++--------
>> 1 file changed, 3 insertions(+), 8 deletions(-)
>> diff --git a/sim/common/sim-arange.c b/sim/common/sim-arange.c
>> index 6373b742ce8..b3488ab564a 100644
>> --- a/sim/common/sim-arange.c
>> +++ b/sim/common/sim-arange.c
>> @@ -32,10 +32,7 @@ along with this program.  If not, see
>> <http://www.gnu.org/licenses/>.  */
>> #include <string.h>
>> #endif
>> -#define DEFINE_INLINE_P (! defined (SIM_ARANGE_C_INCLUDED))
>> -#define DEFINE_NON_INLINE_P defined (SIM_ARANGE_C_INCLUDED)
>> -
>> -#if DEFINE_NON_INLINE_P
>> +#ifdef SIM_ARANGE_C_INCLUDED
>> /* Insert a range.  */
>> @@ -280,9 +277,7 @@ sim_addr_range_delete (ADDR_RANGE *ar,
>> address_word start, address_word end)
>>   build_search_tree (ar);
>> }
>> -#endif /* DEFINE_NON_INLINE_P */
>> -
>> -#if DEFINE_INLINE_P
>> +#else /* SIM_ARANGE_C_INCLUDED */
>> SIM_ARANGE_INLINE int
>> sim_addr_range_hit_p (ADDR_RANGE *ar, address_word addr)
>> @@ -301,4 +296,4 @@ sim_addr_range_hit_p (ADDR_RANGE *ar, address_word addr)
>>   return 0;
>> }
>> -#endif /* DEFINE_INLINE_P */
>> +#endif /* SIM_ARANGE_C_INCLUDED */
> 
> The patch LGTM, I agree these macros are not really needed.  But I am confused, does this actually fix your c++ build?  You mentioned you were using g++ (and therefore cpp), which supposedly can handle this fine:
> 
> https://gcc.gnu.org/onlinedocs/cpp/Defined.html
> 
> If so, what did the error look like?
> 
> Simon

Re: [PATCH] Do not expand macros to 'defined'

[Andreas Schwab]

On Jan 16 2019, Simon Marchi <simon.marchi@polymtl.ca> wrote:

> The patch LGTM, I agree these macros are not really needed.  But I am
> confused, does this actually fix your c++ build?  You mentioned you were
> using g++ (and therefore cpp), which supposedly can handle this fine:
>
> https://gcc.gnu.org/onlinedocs/cpp/Defined.html
>
> If so, what did the error look like?

This has nothing to do with C vs. C++, both have the same wording:

    If the token defined is generated as a result of this replacement
    process or use of the defined unary operator does not match one of
    the two specified forms prior to macro replacement, the behavior is
    undefined.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Re: [PATCH] Do not expand macros to 'defined'

[Simon Marchi]

On 2019-01-16 15:48, Andreas Schwab wrote:
> On Jan 16 2019, Simon Marchi <simon.marchi@polymtl.ca> wrote:
> 
>> The patch LGTM, I agree these macros are not really needed.  But I am
>> confused, does this actually fix your c++ build?  You mentioned you 
>> were
>> using g++ (and therefore cpp), which supposedly can handle this fine:
>> 
>> https://gcc.gnu.org/onlinedocs/cpp/Defined.html
>> 
>> If so, what did the error look like?
> 
> This has nothing to do with C vs. C++, both have the same wording:
> 
>     If the token defined is generated as a result of this replacement
>     process or use of the defined unary operator does not match one of
>     the two specified forms prior to macro replacement, the behavior is
>     undefined.
> 
> Andreas.

That's why I was confused, the purpose of Pavel's patch that was 
replaced by this one is C++-oriented:

https://sourceware.org/ml/gdb-patches/2019-01/msg00241.html

So in the end it looks like the issue is indeed not about C++, but about 
his usage of -Wextra (-Wexpansion-to-defined).

Simon

Re: [PATCH] Do not expand macros to 'defined'

[Simon Marchi]

On 2019-01-16 15:38, Pavel Kryukov wrote:
>> which supposedly can handle this fine
> 
> Right, but not with '-Wall -Wextra -Werror' flags.
> These pedantic options help us (a little) keeping C++ project portable
> between compilers, although we do not intent to have this property on
> GDB integration.

Thanks, that clears it up.  I pushed the patch.

Simon