[v3] libstdc++/12239

[Mike Stump]

On Wednesday, September 17, 2003, at 09:09 PM, Phil Edwards wrote:
>>> default to the most popular, or latest version.
>
> Bleh, so we can silently guess wrong?

If two people disagree with what it should do, then one of them will 
label the behavior wrong.  We can accept their label, and fix it, but 
then the other, unless they change their opinion, will label it wrong.  
Yes, you cannot win in that case.  Solution, don't accept their claim 
it is wrong, and document what it does, if you care, and call the 
behavior right by definition.

This is the status quo and has been for more than 10 years.  I'm not 
trying to say anything controversial.

> Nobody has any idea what the "most popular" version is

Sure we do.

> and defaulting to the latest would I believe be incorrect for most of 
> the developers on this list (for example).

Fine, great, but that would mean you have an idea!  If you think 2.9 is 
the right answer, than that's the answer, if 2.9 is wrong, and 2.8 is 
right, then fine, that can be the right answer.  In the end, the people 
that care will express themselves, and that will be the default.

> If it were just matter of not producing optimal code, then guessing
> wrong might be an acceptable mistake, depending on who you ask.  But
> failing to compile for no obvious reason, or compiling the library and
> causing incorrect data at runtime, is not acceptable.

?  Failing to compile when users ask for things that cannot work, isn't 
a bug.  Next issue.  The error message the software gives can be made 
arbitrarily good, we welcome all such patches.

I just don't get the point you are trying to make.  I don't think it is 
a very interesting point, no matter.  Native builds should just omit 
triplets, and when they do, it just works.  Giving a triplet implies a 
cross compiler, and then, the triplet has to match the provided headers 
and libraries, and failure to match them is a user error and 
uninteresting.

If you think there is a bigger issue here, please provide links to 
mailing list reported problems from the last decade in which we have 
been doing this, and a design of how we can improve the situation.  
Personally, I don't recall seeing more than a few instances of this, 
and each of them were quickly corrected.

?