Re: [patch] std::string::operator[] extension in debug mode

[Douglas Gregor <doug dot gregor at gmail dot com>]

On May 19, 2005, at 6:04 AM, Paolo Carlini wrote:
> Thanks Jonathan for working on this and explaining in public the whole
> thing! My plan would be returning to it in a couple of days, after
> Chris' last batch of move semantics and the visibility long standing 
> PR,
> already in my queue. My impression is that changing the build options 
> is
> indeed unfortunately necessary, because, basically, string and wstring
> are instantiated in the library (actually, I seem to remember that, in
> order to work a bit on the operator[] issue without rebuilding I used
> basic_string<short>). We have to carefully document this, in case.
>
> But better also involving Doug, if possible. Doug can you possibly tell
> us your take on this issue?

My knee-jerk reaction would be to do as Jon has said: turn off the 
extern templates in debug mode. It's an ODR violation, but in a sense 
the entire debug mode is one big ODR violation: we've taken names like 
std::vector and rerouted them to totally different implementations. The 
ODR violation doesn't cause problems because of strong using; we just 
need to ensure that turning off extern templates does the right thing. 
We might also want to mark operator[] with the "always_inline" 
attribute, so that we never get two conflicting definitions of the same 
code.

	Doug