This is the mail archive of the mailing list for the libstdc++ project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: PR3042

>>>>> "David" == David Edelsohn <> writes:

    David> "If it ain't broken, don't fix it."

That is a good argument.  

I am glad that you now understand the situation better. That allows us
to have a far more coherent discussion.

Your argument is precisely the reason that I am supportive of your
efforts to use -brtl.  If we were starting from scratch, with no
implementation of G++ for a system with COMMON, I would be arguing
much more strongly for my change.  The fact that the current behavior
is the historical behavior is the only thing I like about it.

As I've said many times, I'm happy to postpone this debate
indefinitely, if you check in your patch.  Perhaps long enough that
systems that have COMMON, but not COMDAT/WEAK, are no longer an issue.


For the record, there really is nothing harder about working with G++
-- after my change -- than before.  The procedure as a developer is
the same: you write your code, you look at the undefined symbols, and
you write explicit instantiations for them.  You just have to write
more.  The real fix is neither yours nor mine -- it is -frepo (which
is what I would call a prelinker) or a template repository (which is
different from -frepo.)

My fix chooses one tradeoff between fragility and the number of
explicit instantiations you have to write.  Current g++ chooses a
different tradeoff: more fragility, less work.

In general, I think am more focused on providing users with simple
models, rather than complex models, even if the complex models are in
some ways more powerful.  I miss Burley, because he often used to make
the same kinds of arguments.  I think that my perspective from more
experience dealing with "average" users, rather than the highly
skilled developers that make up the traditional free-software base,
than some folks.  But, obviously, my biases are just that, and only
time will tell whether I can be persuasive.

I have one convert, on this particular issue: Benjamin has indicated
that he would find it easier to work on V3 with my change; obviously,
the KDE people might feel differently.

Let's drop the debate at this point.

I'm sure everyone else is entirely bored, and we do have a release to

Mark Mitchell         
CodeSourcery, LLC     

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]