Re: Safety of subprogram'unchecked_access

[stt@houdini.camb.inmet.com (Tucker Taft)]

dennison@telepath.com wrote:

: I have an arcane question about using 'access on procedures.

: I have the following situation: a generic package's body instantiates another
: generic package. The other generic package has a routine that takes an
: "access function" parameter. The "outer" generic's body passes into that
: routine the 'access of a function declared in its body. When I try to compile
: the "outer" generic's body, I get an error on that call stating
: "LRM:3.10.2(32), If the subprogram denoted by the prefix is declared within a
: generic unit, the expected type must be declared within that same generic
: unit, Continuing"

: That is indeed the gist of 3.10.2(32). I can see where there could be a
: problem with the generic owner of the routine going out of scope. But it
: seems that 'Access would be safe in my situation. The "access function" type
: is declared in a package that cannot outlive the owner of the routine who's
: address is passed in.

This restriction is not related to lifetime.  It has to do with the
possibility that subprograms declared in a generic body might require
extra "context" when called, if the implementation uses generic code
sharing.  For implementations that do strictly "macro"-like generic
code expansion, this restriction is not useful.  But portability dictated
that the restriction be independent of implementation approach.

: 'Unchecked_Access is not allowed for subprograms, so how do I get around
: this? The only solution I can see is to change the routine into a generic
: parameter to the "inner" generic. But that loses the ability to change it on
: the fly. Yuk.

You can move the instantiation of the generic, or the declaration
of the subprogram, so they are both in the spec, or both in the body.

: --
: T.E.D.

--
-Tucker Taft   stt@inmet.com   http://www.inmet.com/~stt/
Intermetrics, Inc.  Burlington, MA  USA