Re: How to get Ada to ?cross the chasm??

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

"Paul Rubin" <no.email@nospam.invalid> wrote in message 
news:87vac4z2lh.fsf@nightsong.com...
> "Randy Brukardt" <randy@rrsoftware.com> writes:
>> All of the Ada containers have "cursors", which serve the reference
>> function. You need something like that to specify where to insert 
>> something
>> in a tree ("insert_child (tree, parent)" - the parent is represented by a
>> cursor).
>
> Ok, but can you insert the same child tree into multiple places in the
> parent?

No, of course not. Because if you could, you'd no longer have a tree. If you 
have shared nodes, then you have a DAG (directed acyclic graph). We talked 
about but ultimately decided not to have a graph container, mainly because 
it is hard to describe how many of the operations would work, particularly 
the iterators. (The iterators have to be able to visit *all* of the nodes 
exactly once in some defined order.)

A DAG or worse yet, an unstructured graph with circularities, is precisely 
the kind of thing I was thinking of when I was talking about "spaghetti 
data".

Data structures should be - ahem - structured! Perhaps they might have some 
weak references (like a cursor) into them, but hopefully not.

...
> What if you want one of the elements of a tree node to be a pointer to
> its parent?

That's a basic part of the tree data structure, every node has a Parent 
operation.

...
> So that's a natural source of circular structure in data.

Nope. You are confusing the references needed to construct a data structure 
with the much weaker ones needed to navigate it. Just because there is a 
mechanism to find the parent doesn't make the structure circular.
Note that as a container, all of the operations on the container keep it 
properly consistent, so if a node is deleted, the parent pointers and the 
like are updated. You (the user of the tree container) do not have to do any 
of this bookkeeping, and you don't have to worry about a broken tree, 
either.

When one builds their own data structures from scratch, one has to worry 
about (accidentally or on purpose) making circularities, about maintaining 
the structure's invariants, as well as memory management. GC might help with 
the memory management end, but a better solution is to let the 
compiler/runtime deal with the entire set of issues by using a pre-made data 
structure. (And you get better syntax and capabilities with the pre-made 
structure, including iterators and (in Ada 2020) aggregates.)

...
>> Depends on the job, of course. Much of my software indeed works that way,
>> because there is much less security issues with compiled-in settings.
>
> There is something to be said for that, but I've still been taught that
> a principle of configuration management (CM) is to keep stuff like
> compilers away from the system as much as you can, so who knows.

Depends on the audience, surely. In programs that I expected to be used by 
less-technical others, one has to provide configuration through a GUI, which 
means that one has to write the GUI, read the results in the main server, be 
prepared to change the configuration on the fly, deal with the tasking 
issues that occur when using variables that might change, etc. A whole heck 
of a lot more work than declaring ...

     package AS_Params is
          Root_Path : constant String := "D:\Webroot\";
          ...
     end As_Params;

This puts the configuration into a single Ada package, so it's not 
distributed everywhere (THAT would be bad). It's easy to find if something 
needs to be changed, and I'm not afraid of compilers (I write 'em as part of 
my living! :-).


> In the
> case of the corporate jargon generator though, I'd consider the phrase
> lists to be data rather than settings.  So having them compiled into the
> program still seems awfully primitive.

See above. I'd probably put all of the phrases into a giant aggregate in a 
configuration package. If someone more ambitious than me wanted to make the 
GUI and other changes necessary to make the things changable on the fly, 
they could do so without altering the rest of the program.

                         Randy.