Re: String_Access in unbounded string handling?

["Randy Brukardt" <randy@rrsoftware.com>]

"Tucker Taft" <tucker.taft@gmail.com> wrote in message 
news:afd791fa-853f-48fa-9223-759b12d4ed87n@googlegroups.com...
On Sunday, January 14, 2024 at 6:05:45?AM UTC-5, Blady wrote:
>> Hello,
>>
>> String_Access is defined in A.4.5 Unbounded-Length String Handling:
>> 7 type String_Access is access all String;
>>
>> and note:
>> 75 The type String_Access provides a (nonprivate) access type for
>> explicit processing of unbounded-length strings.
>>
>> I wonder what String_Access is for and what could be "explicit 
>> processing"?
>
>The idea was to support the explicit use of new String'(...), X.all, and 
>Unchecked_Deallocation
>rather than the implicit use of the heap inherent in Unbounded strings.  It 
>was recognized that you
>need a single global access type to avoid having to do conversions all over 
>the place.  This
>predated the availability of stand-alone objects of an anonymous access 
>type
>(aka "SAOOAAATs" ;-), but those are not universally loved either.  It 
>certainly cannot be
>removed now without potentially very painful disruption of existing users. 
>It could be moved
>to a different package without too much disruption, but I haven't seen any 
>groundswell of interest
>in doing that either.

I'm dubious that there are any such users. Certainly, in the handful of 
cases where I needed such a type, I just declared it (strong typing, you 
know?) and never thought of Ada.Strings.Unbounded as being a place to find 
such a type already defined. It is such an odd place I doubt anyone outside 
of perhaps the people who defined the type ever used it.

OTOH, I agree that the compatibility impact is non-zero (anyone who did use 
it would have to change their code), and the benefit of removing the type at 
this point is close to zero (junk declarations abound in long-term Ada 
packages, what's one more; and certainly there is a lot of unused stuff in 
any particular reusable package and any particular use), so the cost-benefit 
ratio doesn't seem to make a change here worth it. An Ada successor language 
would design Ada.Strings.Unbounded rather differently (so as to be able to 
use string literals directly with the type) and probably would include 
universal character support as well, so it's hard to find an important 
reason to change this.

Also, I'm pretty sure we're discussed this within the ARG several times in 
the past, so this is well-trodden ground.

                  Randy.