Re: Best Ref-counting algorithms?

George Neuner schrieb:


Thanks for your detailed explanations [snipped].


> Of course, none of this helps if you don't have the source or the
> meta-information - such as with a canned library. However, in that
> case, call chain analysis will fail to reach the leaves of the call
> tree and so will have to assume the value(s) in question escape(s).


I've been thinking about pointers and references in the last days.
Dynamic runtime behaviour may deserve similar dynamic support for
tracking/limiting the use of pointers. As long as (canned) libraries
have attributed entry points, created by the compiler during their
creation, the same compiler can trust that information.


In other cases it can be helpful, at least, to distinguish between kind
of managed and unmanaged code (see .NET), so that the compiler can know
what a called function is allowed to do with the passed arguments. The
key point is cooperation of the involved tools (compiler, librarian,
linker...) and their use of (binary) file formats with room for required
attributes. The C++ approach with mangled names is interesting, but it
reveals pro's and con's at the same time.


DoDi

Return to the comp.compilers page.
Search the comp.compilers archives again.