Re: [Valgrind-developers] Matching core and skin versions

[Jeremy F. <je...@go...>]

On Wed, 2002-11-06 at 07:01, Nicholas Nethercote wrote:
> Yeah... although I can't help the feeling that there should be a way of
> doing it so that the skin-writer doesn't have to do anything more than
> #include vg_skin.h... bleh.

Well, the skin writer has to do some syntactic mucking about to set up
all the skin entrypoints, so a single extra line isn't going to hurt too
much.

> Unless... if a variable is defined as __attribute__ ((weak)), it can be
> defined by multiple .o files and the linker won't complain, so long as an
> initial value is provided... it just uses whichever definition comes
> first.  All the definitions would be the same because they all arise from
> #include'ing vg_skin, so this should be ok.  I think.  Unless there are
> other complications that I'm not aware of (almost undoubtedly).

I think that's trying to hide too much clever stuff.  For example, its
another thing to go wrong in subtle ways if the various .o files get
different definitions.

> I guess one danger of this would be that the skin-writer could overwrite
> the variables with his own strong versions, but then again the skin-writer
> can already do a million stupid things that will screw up the core so that
> doesn't make much difference.

Given the number of other things a skin writer has to initialize and/or
register, a single extra one for interface version is not a huge
burden.  Far better than making things to clever-clever which becomes
confusing if you're trying to track down a problem later.

	J