Re: [RFC] type safe trees

[Daniel Berlin <dberlin at dberlin dot org>]

On Jun 24, 2004, at 10:34 PM, Ian Lance Taylor wrote:

>  STL is obviously useful for vectors, among
> other things--and we are already using C++, so why not use STL?
> So sooner or later we are using everything except multiple
> inheritance, and the warning option for the parts of C++ which we are
> not supposed to use becomes rather less useful.
>
> I'll add that while I think that that is what would happen, I don't
> think that is necessarily bad.  C++ lets you generate drastically
> inefficient code much more quickly than C, but I think we have the
> infrastructure to detect those problems and fix them.
Yes we do.  Automatic compile time regression whining is not something 
that is even all that difficult, it's just something we need to commit 
to doing.
We should set up reversion policy for compile time regressions that is 
harsher than what we have now.

>  And I think that C++ has clear advantages over C.
Of course.
And inefficient code is (of course) about the programmers, not the 
language.
Vigilance is what makes your compiler faster, not writing in C.
Haven't we proved by example that C is no harder to make a compiler 
slow down in than any other language is?
The simple fact is that if we can't prevent a C++ compiler from slowing 
down, we aren't going to prevent a C one from doing it either, it just 
might happen at a slower rate.
Sad, but true.
Luckily I'm an optimist, and i think our current trend of getting 
slower can be reversed.

--Dan