Re: [rfa/dwarf] Use objfile_data mechanism for per-objfile data

[Daniel Jacobowitz <drow@false.org>]

On Thu, Apr 01, 2004 at 03:57:16PM -0500, Jim Blandy wrote:
> 
> Wonderful!  So it turns out 'struct dwarf2_pinfo' has only one real
> member, huh?  :)

On my branch it has two - I added a per-psymtab hash table in some
cases.

> I have to say, 'dwarf2_per_objfile_data', 'struct
> dwarf2_per_objfile_data', and 'dwarf2_per_objfile' are not my favorite
> cluster of names.  It took me a few passes to get it straight.  (Yes,
> I should have gotten more sleep, but I suspect there are others who
> work under the same conditions...)
> 
> How about:
> - 'dwarf2_objfile_data_key' for the 'struct objfile_data', and
> - 'struct dwarf2_objfile' and 'dwarf2_objfile' for the actual
>   per-objfile datatype and the global pointer to the current instance?
> 
> (Is that any better?  I think suffixes like "_data" really only belong
> on things whose type is unspecified at the point where the name
> appears, like 'void *' pointers, or objects related to them.  I mean,
> everything is "data"; if you're going to give something a
> generic-sounding name, that should be because you're emphasizing the
> genericness of it.)

I don't think that's any better.  "dwarf2_objfile" implies that it's a
kind of objfile.  But I'm willing to use your names :)

> The lower-case implicit-parameter macros bug me.  But I assume they're
> going away soon, and upper-casing them would make the patch huge,
> right?

I didn't have a particular plan in either direction.  Doing either
would be an easy follow-on.  Replacing them with their expansions would
be noisy indentation-wise, but otherwise trivial - that may be best.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer