Pierre THIERRY wrote:

Le Sat, 17 Jun 2006 18:53:19 -0400, David Abrahams a écrit :

...

Usually, in the debugger, you cast the base address of the actual storage to a pointer to the type stored, and go from there.

Thanks a lot, it did the trick! In fact, I immediately ran into problems about namespaces, but I managed to find in recent gdb newsgroups postings how to do...

So now I can debug my varaints! Maybe that could be valuable if such an information was in the Boost documentation?

What Dave suggested is standard information for anyone using a debugger in C++. Putting that in the Boost documentation is unnecessary.

Pierre THIERRY writes:

Le Sun, 18 Jun 2006 05:31:53 -0400, Edward Diener a écrit :

...

What Dave suggested is standard information for anyone using a

...

debugger in C++. Putting that in the Boost documentation is

...

unnecessary.

Without using aligned storage, I do not see why anyone would have to

do such a cast to debug data structures... So it doesn't seem to me

to be standard in any way for C++.

I assume unions have the same problem. So does shared_ptr<void>. And void*. All that said, I'm sympathetic to your point of view.

BTW, is there any way for the debugger to know automatically what the

type the storage should be cast to?

Not without some extra-linguistic mechanism to tell it what type is stored. It would have to use some sort of debugger-specific hook. -- Dave Abrahams Boost Consulting www.boost-consulting.com

David Abrahams wrote:

Edward Diener writes:

...

Pierre THIERRY wrote:

...

...

Le Sat, 17 Jun 2006 18:53:19 -0400, David Abrahams a écrit :

...

...

...

Usually, in the debugger, you cast the base address of the actual

...

...

...

storage to a pointer to the type stored, and go from there.

...

...

Thanks a lot, it did the trick! In fact, I immediately ran into problems

...

...

about namespaces, but I managed to find in recent gdb newsgroups

...

...

postings how to do...

...

...

So now I can debug my varaints! Maybe that could be valuable if such an

...

...

information was in the Boost documentation?

...

What Dave suggested is standard information for anyone using a debugger

...

in C++. Putting that in the Boost documentation is unnecessary.

I disagree. Certainly Pierre's experience belies that statement.

More generally, variant presents a high-level abstraction, and users

aren't necessarily expected to have any idea how it works internally.

For the most part, a naive user with a good debugger can explore his

program's data structures at will, but unlike almost any other library

component, variant is a barrier to the type information the debugger

needs.

Variant isn't the only data structure where the actual type stored is not known at compile time and therefore unavailable to the debugger to show to the user. You yourself pointed out the others. Every good programmer of which I know is cognizant of the technique of casting the variable or memory area, from within a debugger, to the actual type so that the debugger shows the value. Are you saying it should be up to Boost to tell the programmer about this technique ? I am not saying it might not be helpful, but I do not think it belongs in the documentation as a general rule since the technique is universal to other similar situations, and not to just one or two Boost libraries.

Edward Diener writes:

...

I disagree. Certainly Pierre's experience belies that statement. More generally, variant presents a high-level abstraction, and users aren't necessarily expected to have any idea how it works internally. For the most part, a naive user with a good debugger can explore his program's data structures at will, but unlike almost any other library component, variant is a barrier to the type information the debugger needs.

Variant isn't the only data structure where the actual type stored is not known at compile time and therefore unavailable to the debugger to show to the user. You yourself pointed out the others.

Yes, there are a very few others. And I was wrong about shared_ptr<void>, since the deleter still has the static type info.

Every good programmer of which I know is cognizant of the technique of casting the variable or memory area, from within a debugger, to the actual type so that the debugger shows the value.

So does that make Pierre a bad programmer? I don't think anyone's programming skill should be indicted based on ignorance of a single debugging hack.

Are you saying it should be up to Boost to tell the programmer about this technique ?

No, I'm saying that maybe Boost should tell the user enough about the implementation of variant so that he knows to look at the address of the aligned storage for the type stored there when debugging. While we're at it, I think it would be courteous to mention the cast technique.

I am not saying it might not be helpful, but I do not think it belongs in the documentation as a general rule since the technique is universal to other similar situations, and not to just one or two Boost libraries.

The technique may be universal, but I think the _problem_ is particular to just one or two Boost libraries. -- Dave Abrahams Boost Consulting www.boost-consulting.com