minmax.h: Use auto for variables in __minmax_array()

[PATCH next] minmax.h: Use auto for variables in __minmax_array()

Posted by david.laight.linux@gmail.com 13 hours ago

From: David Laight <david.laight.linux@gmail.com>

While 'auto __element = _array[--__len]' should remove 'const',
gcc prior to version 11 are buggy and retain it.
However forcing an integer promotion by adding zero does work.

Promoting signed/unsigned char and short to int doesn't matter here,
that happens as soon as the value is used.

Type type of the result (for char/short arrays) changes, but the value
will always be promoted to int before it is used (for any purpose) so
it isn't even worth casting the type back - all that is likely to do
is make the compiler explicitly mask it to 8/16 bits before it is
immediately promoted back to int.

Signed-off-by: David Laight <david.laight.linux@gmail.com>
---
 include/linux/minmax.h | 19 ++++++-------------
 1 file changed, 6 insertions(+), 13 deletions(-)

diff --git a/include/linux/minmax.h b/include/linux/minmax.h
index a0158db54a04..7d437f73a6d6 100644
--- a/include/linux/minmax.h
+++ b/include/linux/minmax.h
@@ -239,20 +239,13 @@
  * ...
  * min = min_array(buff, nb_items);
  * --- 8< ---
- *
- * The first typeof(&(array)[0]) is needed in order to support arrays of both
- * 'int *buff' and 'int buff[N]' types.
- *
- * The array can be an array of const items.
- * typeof() keeps the const qualifier. Use __unqual_scalar_typeof() in order
- * to discard the const qualifier for the __element variable.
  */
-#define __minmax_array(op, array, len) ({				\
-	typeof(&(array)[0]) __array = (array);				\
-	typeof(len) __len = (len);					\
-	__unqual_scalar_typeof(__array[0]) __element = __array[--__len];\
-	while (__len--)							\
-		__element = op(__element, __array[__len]);		\
+#define __minmax_array(op, array, len) ({			\
+	auto __array = &(array)[0];				\
+	auto __len = len;					\
+	auto __element = __array[--__len] + 0;			\
+	while (__len--)						\
+		__element = op(__element, __array[__len]);	\
 	__element; })
 
 /**
-- 
2.39.5

Re: [PATCH next] minmax.h: Use auto for variables in __minmax_array()

Posted by David Laight an hour ago

On Fri,  6 Feb 2026 22:25:54 +0000
david.laight.linux@gmail.com wrote:

Cc the people discussing unqual_scalar_typeof() for arm64 LTO READ_ONCE().


> From: David Laight <david.laight.linux@gmail.com>
> 
> While 'auto __element = _array[--__len]' should remove 'const',
> gcc prior to version 11 are buggy and retain it.
> However forcing an integer promotion by adding zero does work.
> 
> Promoting signed/unsigned char and short to int doesn't matter here,
> that happens as soon as the value is used.
> 
> Type type of the result (for char/short arrays) changes, but the value
> will always be promoted to int before it is used (for any purpose) so
> it isn't even worth casting the type back - all that is likely to do
> is make the compiler explicitly mask it to 8/16 bits before it is
> immediately promoted back to int.
> 
> Signed-off-by: David Laight <david.laight.linux@gmail.com>
> ---
>  include/linux/minmax.h | 19 ++++++-------------
>  1 file changed, 6 insertions(+), 13 deletions(-)
> 
> diff --git a/include/linux/minmax.h b/include/linux/minmax.h
> index a0158db54a04..7d437f73a6d6 100644
> --- a/include/linux/minmax.h
> +++ b/include/linux/minmax.h
> @@ -239,20 +239,13 @@
>   * ...
>   * min = min_array(buff, nb_items);
>   * --- 8< ---
> - *
> - * The first typeof(&(array)[0]) is needed in order to support arrays of both
> - * 'int *buff' and 'int buff[N]' types.
> - *
> - * The array can be an array of const items.
> - * typeof() keeps the const qualifier. Use __unqual_scalar_typeof() in order
> - * to discard the const qualifier for the __element variable.
>   */
> -#define __minmax_array(op, array, len) ({				\
> -	typeof(&(array)[0]) __array = (array);				\
> -	typeof(len) __len = (len);					\
> -	__unqual_scalar_typeof(__array[0]) __element = __array[--__len];\
> -	while (__len--)							\
> -		__element = op(__element, __array[__len]);		\
> +#define __minmax_array(op, array, len) ({			\
> +	auto __array = &(array)[0];				\
> +	auto __len = len;					\
> +	auto __element = __array[--__len] + 0;			\
> +	while (__len--)						\
> +		__element = op(__element, __array[__len]);	\
>  	__element; })
>  
>  /**

Re: [PATCH next] minmax.h: Use auto for variables in __minmax_array()

Posted by Andrew Morton 13 hours ago

On Fri,  6 Feb 2026 22:25:54 +0000 david.laight.linux@gmail.com wrote:

> From: David Laight <david.laight.linux@gmail.com>
> 
> While 'auto __element = _array[--__len]' should remove 'const',
> gcc prior to version 11 are buggy and retain it.

With what effect?

> However forcing an integer promotion by adding zero does work.
> 
> Promoting signed/unsigned char and short to int doesn't matter here,
> that happens as soon as the value is used.
> 
> Type type of the result (for char/short arrays) changes, but the value

s/Type type/Type/ ?

> will always be promoted to int before it is used (for any purpose) so
> it isn't even worth casting the type back - all that is likely to do
> is make the compiler explicitly mask it to 8/16 bits before it is
> immediately promoted back to int.

I'm not understanding the motivation for this change.  Is there some
compilation issue to be addressed?

Re: [PATCH next] minmax.h: Use auto for variables in __minmax_array()

Posted by David Laight an hour ago

On Fri, 6 Feb 2026 14:41:35 -0800
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Fri,  6 Feb 2026 22:25:54 +0000 david.laight.linux@gmail.com wrote:
> 
> > From: David Laight <david.laight.linux@gmail.com>
> > 
> > While 'auto __element = _array[--__len]' should remove 'const',
> > gcc prior to version 11 are buggy and retain it.  
> 
> With what effect?

If you have:
int f(const int x)
{
	auto y = x;
	y++;
	return y;
}
gcc prior to 11.0 error that y is const.
So in this case the loop can't change __element.

The constness is also kept by 'auto y = +x' which does integer promotion
(useful for converting enums) and both -x and ~x.

> > However forcing an integer promotion by adding zero does work.
> > 
> > Promoting signed/unsigned char and short to int doesn't matter here,
> > that happens as soon as the value is used.
> > 
> > Type type of the result (for char/short arrays) changes, but the value  
> 
> s/Type type/Type/ ?

Actually s/Type/the/

> > will always be promoted to int before it is used (for any purpose) so
> > it isn't even worth casting the type back - all that is likely to do
> > is make the compiler explicitly mask it to 8/16 bits before it is
> > immediately promoted back to int.  
> 
> I'm not understanding the motivation for this change.  Is there some
> compilation issue to be addressed?

Mainly unqual_scalar_typeof() being horrid.
There is an ongoing long thread about its use in the arm64 LTO READ_ONCE().
Newer compilers do have a builtin, and there are some shorter alternatives
that work in some places.
But here is just isn't needed.
So one less place to check.

I did mean to copy the main contributers to that thread, but forgot.

	David