Re: [CEDET-devel] Changes to c.by for parsing 'using' and 'namespace'

[Eric M. L. <er...@si...>]

On Thu, 2009-08-13 at 15:46 +0200, David Engster wrote:
> Eric M. Ludlam <er...@si...> writes:
> >   I applied your patch, but had a test failure from the parsed output.
> > I had to fiddle with it for a while to get everything working again.
> >
> >   The original code creating 'using tags with a :type that was itself a
> > tag with a :prototype set to t.  What I checked in keeps that
> > distinction, while adding differentiation between classes and
> > namespaces.
> 
> Thanks for correcting this. I do not really understand why the original
> patch broke some tests; I made sure that 'make utest' still works. Do
> you have additional tests on your machine which are not in CVS?

I only had a patch from you I had integrated and checked in on July 17.

> >   I also added your alias item.  The typedef equivalent of an alias does
> > not use a :kind meaning an alias, it's just a typedef tag.  It then
> > needed special code to use the typedef as an alias while looking up
> > symbols.  This new thing is a bit different though, so I didn't mess
> > with it.
> 
> If you prefer to use a typedef tag for a C++ namespace alias, I guess I
> could change that. 
Tags should always describe the code as best it can, so I think it
should be a namespace with attributes to describe what it does.  We'll
just need to add special code to interpret this tag.

> The reason I did it this way is to resolve all kinds
> of namespace aliasing in one dereferencer (which I haven't posted
> yet). This dereferencer tries to resolve stuff like
> 
> namespace foo = bar;
> 
> as well as, for example,
> 
> namespace foo {
>   using namespace bar;
> }
> 
> Both statements practically make 'foo' an alias for 'bar', so I tried to
> parse them in a similar manner. (The latter statement has some
> additional consequences regarding symbol lookup in namespace 'foo', but
> that's another matter.)

C++ is full of fun tricks.

> >   I've forgotten a bunch from this thread.  Where there new tests
> > related to this patch that uses the namespace/alias thing?
> 
> No. These patches to c.by still need the namespace-dereferencer (which
> is finished; I'll post it when we have figured out the 'fully qualified
> parent' stuff, since it depends a little bit on how that will be
> implemented in the end).

I still think that the name lookup of some parent class should be scoped
to the location of the parent tag.  That was one of the things I wanted
to resolve next.

Eric