Re: [Bug-apl] API for GNU APL

[Juergen Sauermann]

Hi Dirk,

A *Value_P* object is basically a pointer to a *Value* object, pretty 
much what is sometimes
called a smart pointer or a shared pointer. The purpose of a *Value_P* 
is to delete the *Value* it
points to when the last *Value_P *is deleted.

You should never use references to *Value_P *objects. You may use 
references to *const Value &* instead
which can be obtained through *Value_P::getref()*.

A *const Value &* is safe as long as the value exists;
A non-const*Value &* is somewhat dangerous.

A *Value_P *should never be allocated via *new* (and not freed by 
*delete*). It should also not be used in the way
suggested by your PCLONE macro below. Instead you can simply copy 
*Value_P* objects like integers:

*Value_P value(new Value(...));   // a pointer to a new Value object**
**Value_P copy = value;            // a copy of the pointer!*

Also, a *Value* object should never be allocated via *new *except when 
constructing a *Value_P* object from it.

Note that above *copy* points to the /same/ *Value* object as *value* 
because *copy* is a copy of the pointer and
not a copy of *value* itself. If you want a different value that is 
equal to the first then you can use *value->clone() *like this:

*Value_P copy = value->clone(LOC);*   // a (deep) copy of the *value*

This is typically needed when *value* is assigned to a variable and then 
modified or deleted .

At some point in time you may want to issue the APL command *)CHECK *to 
see if your code has created any
stale *Value* objects.

/// Jürgen


On 02/22/2015 10:26 AM, Dirk Laurie wrote:
> Hi Jürgen:
>
> I've got code that works except for bugs unconnected with the GNU APL
> routines.
>
> I decided to implement the interface as a wrapper for an item of type
> `const Value_P&`. API.cc contains the following lines:
>
> // The commented-out variants give segmentation errors, I don't understand why.
> // Maybe needed only when `value` is Value*. not when it is already Value_P.
> // Jürgen will know.
> // #define PCLONE(valuep) valuep
> // #define PCLONE(valuep) Value_P(valuep)
> #define PCLONE(valuep) *(new Value_P(valuep))
>
> I.e. instead of initializing the item as e.g. `symbol->get_value()`,
> I initialize it as `PCLONE(symbol->get_value())`.
>
> I have not laboriously checked in which places this additional `new`
> is really necessary, all I know is that it is necessary for at least one
> of them.
>
> Best wishes
> Dirk