Re: [boost] [serialization] Notes from a new boost user

From: Michael Behrns-Miller (m_at_[hidden])
Date: 2005-07-01 16:51:09

> David Abrahams wrote:
>> "Michael Behrns-Miller" <m_at_[hidden]> writes:
>>
>>> 2) The new "const required to save" restriction bit me, took a day or
>>> two to work through the mailing list and #boost to figure out the
>>> problem.  Is
>>
>> Did it prevent a bug, or did it just cost you 2 days of your valuable
>> time?
>
> Were you not tempted to just use "const_cast"? Why not?

I temporarily tried it and got it to work, but that wasn't an acceptable
solution to me.  I hate casting, immediately makes me reassess the design.
 Which I did, which led me to the answer.  :P

I expected to have ramp-up time with boost.  I probably spent a day
fiddling with bjam, a half day figuring out to include boost libraries,
and a day or two on the const issue.  Not full days, just poking around in
my spare time.  I got everything working under vc71 and gcc 3.3.5 way
before getting frustrated, so I was happy.

Also, to be fair to you, I was using a verbatim copy of the demo code to
start, which I believe needs updating.  I was working from documentation
that didn't cover the const issue.  With those two improvements, I would
have been better off.

But I am not foolish enough to pretend I have the answer to the const
issue, especially given there are many current serialization users.  Just
providing some feedback.  <ducks and runs>

> Whenever I do include a BOOST_STATIC_ASSERT that a library user might be
> expected to trip on, I include comments at that point which describes the
> problem and hopefully a helpful course of action. When I build in my
> environment (VC 7.1) and double click on the error message, the
> BOOST_STATIC_ASSERT pops up and right there is the comment explaining it.
> To me, this seems pretty good and I would hope that this would minimize
> the time a user has to spend looking for sources to these problems.  In
> fact, I have received mail from users that have remarked upon exactly
> this feature/practice as being very helpful in understanding these other
> wise very subtle errors.  By far the best one is when one uses an
> xml_archive to serialize something which doesn't have a variable name
> attached.  It pops a STATIC_ASSERT and the comment right there is the
> comment describing what one is doing wrong.

I agree, that's where I started when figuring it out, looking at the error
line, which clued me in.

Thanks again.