Use using instead of typedef [kdtree, keypoints, octree]

[SunBlack]

Changes are done by: run-clang-tidy -header-filter='.' -checks='-,modernize-use-using' -fix

[taketwo]

Any idea what did this attribute do?

[SergioRAgostinho]

I have the feeling it likely disables aliasing warnings that will be triggered by this.

pcl/keypoints/src/brisk_2d.cpp

Line 1678 in 2f0dc49

const UCHAR_ALIAS* result = reinterpret_cast<const UCHAR_ALIAS*> (&result1);

More on the topic:
https://stackoverflow.com/questions/6313050/gcc-how-to-use-attribute-may-alias-properly-to-avoid-derefencing-type
https://en.wikipedia.org/wiki/Aliasing_(computing)#Conflicts_with_optimization

[taketwo]

Let's maybe leave the typedef here?

[SergioRAgostinho]

I'm surprised the same logic can't be expressed with using.

[taketwo]

Maybe it could. But frankly speaking... I have no desire to study this question. Getting rid of each and every typedef is a no-goal for us, so leaving as-is is the safest option I think.

[taketwo]

@SunBlack according to this page, using alias should look like:

using identifier attr(optional) = type-id 

That is, attributes come before equals sign. Right now attributes are after equals and the thing still compiles, however we don't know whether it works as expected. I'd suggest we keep typedef to avoid potential troubles.

[SunBlack]

You are right, my code is not correct. I played now a bit with it by using a aliasing example from here. I adjusted the code the to use aliasing typedef, see here.

Result:
typedef float FLOAT_ALIAS; and using FLOAT_ALIAS = __attribute__ ((__may_alias__)) float; are producing:

foo(float*, int*): # @foo(float*, int*)
  mov dword ptr [rsi], 1
  mov dword ptr [rdi], 0
  mov eax, 1
  ret

typedef float __attribute__ ((__may_alias__)) FLOAT_ALIAS; and using FLOAT_ALIAS __attribute__ ((__may_alias__)) = float; are producing:

foo(float*, int*): # @foo(float*, int*)
  mov dword ptr [rsi], 1
  mov dword ptr [rdi], 0
  mov eax, dword ptr [rsi]
  ret

@taketwo I believed attr in using identifier attr(optional) = type-id is only for new attributes like [[deprecated]], but are right :).

[SunBlack]

Just to add a note: I don't believe we need aliasing here anywhere, because we are just reading value of result1 (so nothing to optimize there).

[taketwo]

👍 thanks for researching assembly, good to know!

[taketwo]

As per inline comment, according to this page, using alias should look like:

using identifier attr(optional) = type-id 

That is, attributes come before equals sign. Right now attributes are after equals and the thing still compiles, however we don't know whether it works as expected. I'd suggest we keep typedef to avoid potential troubles.