Re: Interfaces.C.Strings chars_ptr memory management strategy

["Alejandro R. Mosteo" <alejandro@mosteo.com>]

On 30/05/2018 21:56, Randy Brukardt wrote:
> "Alejandro R. Mosteo" <alejandro@mosteo.com> wrote in message
> news:pem7s0$go4$1@dont-email.me...
>> On 26/05/2018 00:22, NiGHTS wrote:
>>> I am creating a binding to a C library that requires me to repeat the
>>> same function but with a string parameter that changes on each call. I
>>> don't want to have to keep creating and destroying string memory for all
>>> of these function calls. I would like to create the memory for the string
>>> once, allocate enough space so it doesn't need to grow, and reuse that
>>> memory for the function call every time I need to pass a new string to
>>> it.
>>
>> I'm currently using this:
>>
>> https://github.com/mosteo/cstrings
>>
>> which is not what you want since it allocates on every instance (although
>> on the stack). Still, it might give you some ideas.
>>
>> I'm still unsure if that's 100% guaranteed to be safe; for the experts out
>> here, the question is, in a call to a C function like this:
>>
>> Call_To_C_Function
>>     (Function_That_Returns_A_Limited_Tagged_Type (...)
>>        .Subprogram_That_Returns_A_C_Pointer_To_Data_In_The_Tagged_Type);
>>
>> Is the in-place built limited tagged type guaranteed to live during the
>> call to the C function? (In other words, is the pointer safe (as long as
>> the C side does not make a copy, of course)?
> 
> That depends on the master of the parameter. I believe that the master of a
> parameter is that of the call (each call being it's own master for the
> parameters) -- you'd have to look in 7.6.1 to be sure. So they stay around
> as long as the call.

Might that be 6.4.1? 7 deals with packages (in 2012).

Although it was too dense for me anyway :(

> If that wasn't true, passing an aggregate could have the temporary object
> freed/finalized before the call ended, which would be a disaster.
> 
>> My suspicion is that once the subprogram returns the pointer, the limited
>> type can be optimized away before the call to the C side. It's not what
>> I'm seeing now, but I don't want to depend on an erroneous assumption.
> 
>   Don't think this is a problem in general -- and the result of a function
> does *not* belong to the master of the call but rather the enclosing one
> (else you couldn't use it before it went away). You might be able to get it
> with nested calls:
> 
>       Foo (Bar (Ugh (...))
> 
> The result of Ugh is finalized when the call to Bar ends, so if it is
> somehow in the result of Bar, you could get trouble in Foo. You can avoid
> that by declaring the parameter "aliased" (those belong to the *result* of
> the function, so they stick around longer).

The parameter is actually aliased already. So I hope your impressions are 
right and I'm on firm ground then, a pleasant surprise.

Thanks,
Alex.

> 
>                                                            Randy.
> 
>