close(file); always necessary??

[Frank Heckenbach]

samiam at moorecad.com wrote:

> Hence the reference count stuffs. I did a paper analysis once, I decided
> that:
> 
> 1. To do true automated storage recycling, you have to track each
> pointer
> leaving scope, which means that you have to know where every pointer is.
> 
> 2. Be ready to reference count each pointer, that is, each copy
> operation
> increments the reference count, and each dynamic entry gets a reference
> counter that is initially one for the pointer that is returned.
> 
> 3. Decrement the reference counter each time any pointer leaves scope.
> 
> 4. produce an error if the programmer attempts to free a non-zero
> referred
> pointer.
> 
> 5. The whole thing generates headaches.
> 
> Seriously, it is quite doable with Pascal, but generates a lot of
> overhead.
> Each data structure needs a template for the contained pointers, and
> every data structure leaving scope or freed must be analysed for
> contained
> pointers, and those pointers decrementing reference counts.

This seems correct, and in principle doable. (One additional
well-known problem with reference counting are cyclic data
structures, so you need additional logic to detect and dispose of
them when unused.)

But it's all a lot of work, and some runtime overhead, also memory
overhead, and incompatible with low-level stuff (including manual
memory allocation which might be brushed off as "dirty, don't do
it", but also interfacing with foreign-language libraries which is
quite a common thing).

> Wirth also described this scheme, although I don't think he ever tried
> it.
> I have thought about implementing it primarily to shut people up, i.e.,
> lack of automated storage collection "is a problem with old languages
> like Pascal". It isn't, the price paid for reclaimation overhead is the
> same as Java or other languages,

I'm not sure it's actually the same. I think there are scenarios
where reference counting is more efficient, and others where garbage
collection (i.e., regular cleanup, by starting from known live
structures) is more efficient. Of course, the latter can also be
done in Pascal, but it also a lot of work. It's partly the same as
for reference counting (in both cases, you need to find pointers in
structures), but partly quite different (finding live bindings, and
the actual garbage collection).

> we are primarily burdened by users who
> expect true compiled performance code.

For some of my programs (not most, though), I'm one of those
burdensome users. :-) So if such a thing is ever implemented (note
that I have no plans to do so), I think it should be optional.

Frank

-- 
Frank Heckenbach, f.heckenbach at fh-soft.de, http://fjf.gnu.de/, 7977168E
GPC To-Do list, latest features, fixed bugs:
http://www.gnu-pascal.de/todo.html
GPC download signing key: ACB3 79B2 7EB2 B7A7 EFDE  D101 CD02 4C9D 0FE0 E5E8