Re: Ada Core Technologies and Ada95 Standards

[dewar@cs.nyu.edu (Robert Dewar)]

John published an example of a supposed bug report. But it concerned
pragma Shared, and pragma Shared only has non-null semantics in the
presence of tasking, but there was no task in the program. So this
is clearly not a bug report.

My guess is that you were assuming that pragma Shared should someone
behave reasonably with respect to undefined outside accesses to something
with an address clause. OK, that is a reasonable assumption, but there is
no requirement to this effect in the RM. An implementation is free to
ignore pragma Shared in a program with no tasks (because no observable
semantic difference is possible under these circumstances between a
shared and non-shared implementation of a given variable).

I don't know if the original example had tasks or not. If so, it is possible
that the purpose of the report has been destroyed in the cutting down
process. This often happens. On the other hand if the original program
was expecting pragma Shared to have some effect on non-tasking situations,
then this is an expectation not backed up by the RM.

Remember that the RM is always talking about "as-if" semantics. When it
gives an implementation model, then any implementation model which is
formally semantically equivalent is acceptable.

So you can't appeal to the RM here -- you can talk about what would be
the most useful implementation approach. Certainly GNAT would not ignore
the pragma in this situation, even though it is formally allowed to.