Re: [Boost-users] errors serializing pointers on Fedora Core 5

From: Robert Ramey (ramey_at_[hidden])
Date: 2006-08-25 01:41:41

Dan Bikel wrote:
> Robert Ramey wrote:
>>> Error No. 1 can be fixed by making pFoo a const pointer type, i.e.,
>>> by changing the declaration/assignment statement to be
>>> string *const pFoo = new string("foo");
>>>
>>> I cannot figure out why such a change is necessary, as the
>>> serialization library documentation does not indicate that it is
>>> necessary to make pointers const in order to serialize them (as is
>>> borne out by the demo.cpp code).
>
>> This is explained in the "Rationale" section of the documentation.
>
> Okay, I read through the (long) scenario described in the Rationale
> section, and I guess I agree with this design decision, but in the
> main documentation--the Tutorial section in particular--there should
> not be code snippets that are incorrect.  I'm thinking of the example
> in the "Pointers" section of the tutorial, where the sample code shows
> *non-const* bus_stop pointers being serialized in order to serialize
> "bus_route" objects.  At the very least, you should have correct code
> and a hyperlink to the Rationale section explaining why the const is
> there.

That section of the manual uses the "&" operator rather than the << 
operator.
This operator doesn't enforce the "trap" - in part to give an escape
for those who find the trap unhelpful.

Actually, there should be a trap - leading to a better error message -
for detecting the case you've got here - serialization of a pointer
to a prmitive - which needs special treatment.  But then - we really
should implement concept checking to detect these errors and
display better messages.  - its on my to do list.

>
> Anyway, I don't mean this to sound harsh, because I'm very grateful
> for your prompt answer to my question!
>
> --Dan.