[Tarantool-patches] [PATCH] decimal: fix build with GCC 10

[Tarantool-patches] [PATCH] decimal: fix build with GCC 10

[Serge Petrenko]

GCC 10 produces the following error:
cc1: warning: function may return address of local variable [-Wreturn-local-addr]

Fix it.

Part-of #4966
---
https://github.com/tarantool/tarantool/issues/4966
https://github.com/tarantool/tarantool/tree/sp/gh-4966-decimal-build

 src/lib/core/decimal.c | 8 ++++----
 src/lib/core/decimal.h | 4 ++--
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/lib/core/decimal.c b/src/lib/core/decimal.c
index c44b2f6a2..019c68338 100644
--- a/src/lib/core/decimal.c
+++ b/src/lib/core/decimal.c
@@ -176,24 +176,24 @@ decimal_to_integer(decimal_t *dec)
 	return decimal_check_status(dec, &decimal_context);
 }
 
-decimal_t *
+const decimal_t *
 decimal_to_int64(const decimal_t *dec, int64_t *num)
 {
 	decimal_t d = *dec;
 	if (decimal_to_integer(&d) == NULL)
 		return NULL;
 	*num = decNumberToInt64(&d, &decimal_context);
-	return decimal_check_status(&d, &decimal_context);
+	return decimal_check_status(&d, &decimal_context) != NULL ? dec : NULL;
 }
 
-decimal_t *
+const decimal_t *
 decimal_to_uint64(const decimal_t *dec, uint64_t *num)
 {
 	decimal_t d = *dec;
 	if (decimal_to_integer(&d) == NULL)
 		return NULL;
 	*num = decNumberToUInt64(&d, &decimal_context);
-	return decimal_check_status(&d, &decimal_context);
+	return decimal_check_status(&d, &decimal_context) != NULL ? dec : NULL;
 }
 
 int
diff --git a/src/lib/core/decimal.h b/src/lib/core/decimal.h
index 9a162c73c..e276344d2 100644
--- a/src/lib/core/decimal.h
+++ b/src/lib/core/decimal.h
@@ -110,11 +110,11 @@ decimal_to_string(const decimal_t *dec);
  * \param[out] num - the result
  * @return NULL if \a dec doesn't fit into int64_t
  */
-decimal_t *
+const decimal_t *
 decimal_to_int64(const decimal_t *dec, int64_t *num);
 
 /** \sa decimal_to_int64 */
-decimal_t *
+const decimal_t *
 decimal_to_uint64(const decimal_t *dec, uint64_t *num);
 
 /**
-- 
2.24.3 (Apple Git-128)

Re: [Tarantool-patches] [PATCH] decimal: fix build with GCC 10

[Alexander V. Tikhonov]

Hi Sergey, thanks a lot for the patch, I've successfully checked it
on Fedora 32 and on the full testing process:
  https://gitlab.com/tarantool/tarantool/-/pipelines/153432554

On Fri, Jun 05, 2020 at 05:01:25PM +0300, Serge Petrenko wrote:
> GCC 10 produces the following error:
> cc1: warning: function may return address of local variable [-Wreturn-local-addr]
> 
> Fix it.
> 
> Part-of #4966
> ---
> https://github.com/tarantool/tarantool/issues/4966
> https://github.com/tarantool/tarantool/tree/sp/gh-4966-decimal-build
> 
>  src/lib/core/decimal.c | 8 ++++----
>  src/lib/core/decimal.h | 4 ++--
>  2 files changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/src/lib/core/decimal.c b/src/lib/core/decimal.c
> index c44b2f6a2..019c68338 100644
> --- a/src/lib/core/decimal.c
> +++ b/src/lib/core/decimal.c
> @@ -176,24 +176,24 @@ decimal_to_integer(decimal_t *dec)
>  	return decimal_check_status(dec, &decimal_context);
>  }
>  
> -decimal_t *
> +const decimal_t *
>  decimal_to_int64(const decimal_t *dec, int64_t *num)
>  {
>  	decimal_t d = *dec;
>  	if (decimal_to_integer(&d) == NULL)
>  		return NULL;
>  	*num = decNumberToInt64(&d, &decimal_context);
> -	return decimal_check_status(&d, &decimal_context);
> +	return decimal_check_status(&d, &decimal_context) != NULL ? dec : NULL;
>  }
>  
> -decimal_t *
> +const decimal_t *
>  decimal_to_uint64(const decimal_t *dec, uint64_t *num)
>  {
>  	decimal_t d = *dec;
>  	if (decimal_to_integer(&d) == NULL)
>  		return NULL;
>  	*num = decNumberToUInt64(&d, &decimal_context);
> -	return decimal_check_status(&d, &decimal_context);
> +	return decimal_check_status(&d, &decimal_context) != NULL ? dec : NULL;
>  }
>  
>  int
> diff --git a/src/lib/core/decimal.h b/src/lib/core/decimal.h
> index 9a162c73c..e276344d2 100644
> --- a/src/lib/core/decimal.h
> +++ b/src/lib/core/decimal.h
> @@ -110,11 +110,11 @@ decimal_to_string(const decimal_t *dec);
>   * \param[out] num - the result
>   * @return NULL if \a dec doesn't fit into int64_t
>   */
> -decimal_t *
> +const decimal_t *
>  decimal_to_int64(const decimal_t *dec, int64_t *num);
>  
>  /** \sa decimal_to_int64 */
> -decimal_t *
> +const decimal_t *
>  decimal_to_uint64(const decimal_t *dec, uint64_t *num);
>  
>  /**
> -- 
> 2.24.3 (Apple Git-128)
> 

Re: [Tarantool-patches] [PATCH] decimal: fix build with GCC 10

[Kirill Yukhin]

Hello,

On 05 июн 17:01, Serge Petrenko wrote:
> GCC 10 produces the following error:
> cc1: warning: function may return address of local variable [-Wreturn-local-addr]
> 
> Fix it.
> 
> Part-of #4966
> ---
> https://github.com/tarantool/tarantool/issues/4966
> https://github.com/tarantool/tarantool/tree/sp/gh-4966-decimal-build

LGTM.
I've checked your patch into 2.3, 2.4 and master.

--
Regards, Kirill Yukhin

Re: [Tarantool-patches] [PATCH] decimal: fix build with GCC 10

[Alexander Turenko]

The patch itself is okay.

On Fri, Jun 05, 2020 at 05:01:25PM +0300, Serge Petrenko wrote:
> GCC 10 produces the following error:
> cc1: warning: function may return address of local variable [-Wreturn-local-addr]
> 
> Fix it.

It would be good to mention whether it is a real problem or
false-positive. (The patch is already pushed, so this comment is just
for information.)

As I see, we actually return a data on the stack before the patch, so it
worth to mention the bugfix in the release notes. Kirill, can you add
this?

@ChangeLog

- Fix use-after-scope when converting a decimal number to int64_t or
  uint64_t that may lead to undefined behaviour in tuple comparators and
  so in behaviour of space indices (part of gh-4966).

(Serge, correct me if I'm wrong here.)

BTW, are those hints pesisted for vinyl? Can we get wrong results even
after upgrade?

WBR, Alexander Turenko.

Re: [Tarantool-patches] [PATCH] decimal: fix build with GCC 10

[Serge Petrenko]

08.06.2020 14:22, Alexander Turenko пишет:
> The patch itself is okay.
>
> On Fri, Jun 05, 2020 at 05:01:25PM +0300, Serge Petrenko wrote:
>> GCC 10 produces the following error:
>> cc1: warning: function may return address of local variable [-Wreturn-local-addr]
>>
>> Fix it.
> It would be good to mention whether it is a real problem or
> false-positive. (The patch is already pushed, so this comment is just
> for information.)


It's a false-positive. The return value  is  only  used to test `dec != 
NULL`.

`dec == NULL` means an error.So, no UB here.

I guess no need for a changelog entry then.

>
> As I see, we actually return a data on the stack before the patch, so it
> worth to mention the bugfix in the release notes. Kirill, can you add
> this?
>
> @ChangeLog
>
> - Fix use-after-scope when converting a decimal number to int64_t or
>    uint64_t that may lead to undefined behaviour in tuple comparators and
>    so in behaviour of space indices (part of gh-4966).
>
> (Serge, correct me if I'm wrong here.
>
> BTW, are those hints pesisted for vinyl? Can we get wrong results even
> after upgrade?
>
> WBR, Alexander Turenko.

-- 
Serge Petrenko

Re: [Tarantool-patches] [PATCH] decimal: fix build with GCC 10

[Alexander Turenko]

On Mon, Jun 08, 2020 at 02:30:43PM +0300, Serge Petrenko wrote:
> 
> 08.06.2020 14:22, Alexander Turenko пишет:
> > The patch itself is okay.
> > 
> > On Fri, Jun 05, 2020 at 05:01:25PM +0300, Serge Petrenko wrote:
> > > GCC 10 produces the following error:
> > > cc1: warning: function may return address of local variable [-Wreturn-local-addr]
> > > 
> > > Fix it.
> > It would be good to mention whether it is a real problem or
> > false-positive. (The patch is already pushed, so this comment is just
> > for information.)
> 
> 
> It's a false-positive. The return value  is  only  used to test `dec !=
> NULL`.
> 
> `dec == NULL` means an error.So, no UB here.
> 
> I guess no need for a changelog entry then.

Ouch, I see. Disregard so.