Re: Need a Wild Library Unit Elaboration Example

[Robert A Duff <bobduff@shell01.TheWorld.com>]

csampson@inetworld.net (Charles H. Sampson) writes:

>      I'm preparing a talk on elaboration with the ultimate 
> goal being an understanding of the problems you can get 
> into with library unit elaboration when you make calls 
> between packages.  Most of the emphasis is on the common 
> stuff: The package's elaboration code or an object's 
> initialization using a subprogram from another package.

You should read the GNAT documentation.  There's a very clear
description of elaboration issues in there somewhere -- 
with discussion of when various pragmas are needed and
whatnot.

>      I like to end the talk with three examples that I 
> characterize as pathological; cross-package dependencies 
> you likely wouldn't think of on your first encounter with 
> the subject.  Unfortunately for me, one of my examples has 
> been obsoleted by pragma Elaborate_Body.  I think three 
> examples is just the right number, so I'm asking for your 
> favorite examples of cross-package dependencies to fill 
> out my list again.

How about a dispatching call to a procedure that doesn't
exist yet?  Or something similar with access-to-procedures?

>      Just to avoid duplication, my two remaining examples 
> are elaboration code that calls a protected object entry 
> whose barrier condition uses a function from another 
> package and elaboration code that calls a task entry that 
> has a call to a subprogram from another package in its 
> rendezvous.  (I'm not promoting this as good coding 
> practice, mind you.  I'm just trying to illustrate how 
> hairy things can get, theoretically.)

I don't see why these are pathological.  Unusual, perhaps.
But still a perfectly reasonable combination of features
that ought to be orthogonal.  Define "pathological".

>...Also, as part of 
> the non-pathological stuff, I've covered the elaboration 
> of a statically allocated task's declarative part as part 
> of the elaboration of a containing package.

I think most of my troubles with elaboration have involved generics --
perhaps elaboration of an instance of the generic doesn't do much of
anything, but Elaborate_All doesn't work, because something calls
something else.  And the instance is usually elaborated at library
level.

A key point is that there's no issue unless there's some sort of mutual
recursion going on.

- Bob