Is this a compiler bug ?

Is this a compiler bug ?

[Rod Kay]

Hi all,

    Came across this during a port of the Box2D physics engine.

    It's a generic Stack package using 'ada.Containers.Vectors' to 
implement the stack.

    One generic parameter is the 'initial_Capacity' of the stack, used 
in the 'to_Stack' construction function, via the Vectors 
'reserve_Capacity' procedure.

    In the 'to_Stack' function, the Capacity is reserved correctly but 
in the test program when the stack is created and assigned to a 
variable, the capacity is 0.

    Here is the (very small) source code ...

    https://gist.github.com/charlie5/7b4d863227a510f834c2bfd781dd50ba


    The output I get with GCC 12.2.0 is ...

[rod@orth bug]$ ./stack_bug
to_Stack ~ Initial Capacity: 256
to_Stack ~ Before reserve:   0
to_Stack ~ After reserve:    256
stack_Bug ~ Actual Capacity: 0


Regards.

Re: Is this a compiler bug ?

[Jeffrey R.Carter]

On 2023-03-19 07:17, Rod Kay wrote:
> 
>     In the 'to_Stack' function, the Capacity is reserved correctly but in the 
> test program when the stack is created and assigned to a variable, the capacity 
> is 0.

I think this is acceptable behavior. See ARM A.18.2 (147.19/3, 147.20/3, & 
147.b/3) (http://www.ada-auth.org/standards/aarm12_w_tc1/html/AA-A-18-2.html). 
The first two sections define the behavior of procedure Assign, while the last 
states "Assign(A, B) and A := B behave identically".

Assign (A, B) only changes the capacity of A if A.Capacity < B.Length.

So if the compiler does not use build-in-place for the initialization of the 
variable, then the assignment of the function result should not change the 
capacity of the variable from its (apparent) default of zero (there is, of 
course, no requirement for the capacity of a default-initialized vector).

The discussion of capacities for vectors is only meaningful for a subset of 
possible implementations, so messing with capacities may have no meaningful 
effect at all.

For an unbounded stack based on a linked list (with no concept of capacity) you 
could use PragmARC.Data_Structures.Stacks.Unbounded.Unprotected 
(https://github.com/jrcarter/PragmARC/blob/Ada-12/pragmarc-data_structures-stacks-unbounded-unprotected.ads).

-- 
Jeff Carter
"Since I strongly believe that overpopulation is by
far the greatest problem in the world, this [Soylent
Green] would be my only message movie."
Charleton Heston
123

Re: Is this a compiler bug ?

[Rod Kay]

On 19/3/23 21:33, Jeffrey R.Carter wrote:
> On 2023-03-19 07:17, Rod Kay wrote:
>>
>>     In the 'to_Stack' function, the Capacity is reserved correctly but 
>> in the test program when the stack is created and assigned to a 
>> variable, the capacity is 0.
> 
> I think this is acceptable behavior. See ARM A.18.2 (147.19/3, 147.20/3, 
> & 147.b/3) 
> (http://www.ada-auth.org/standards/aarm12_w_tc1/html/AA-A-18-2.html). 
> The first two sections define the behavior of procedure Assign, while 
> the last states "Assign(A, B) and A := B behave identically".
> 
> Assign (A, B) only changes the capacity of A if A.Capacity < B.Length.
> 
> So if the compiler does not use build-in-place for the initialization of 
> the variable, then the assignment of the function result should not 
> change the capacity of the variable from its (apparent) default of zero 
> (there is, of course, no requirement for the capacity of a 
> default-initialized vector).
> 
> The discussion of capacities for vectors is only meaningful for a subset 
> of possible implementations, so messing with capacities may have no 
> meaningful effect at all.
> 
> For an unbounded stack based on a linked list (with no concept of 
> capacity) you could use 
> PragmARC.Data_Structures.Stacks.Unbounded.Unprotected 
> (https://github.com/jrcarter/PragmARC/blob/Ada-12/pragmarc-data_structures-stacks-unbounded-unprotected.ads).
> 


    Thank you, Jeffrey, for the detailed reply.

    I'm now using a limited record with an extended return for 
'build-in-place' initialisation and am getting the behavior I desired.


Cheers.