Joseph S. Myers wrote:
In the case mentioned of __FUNCTION__ concatenation, (a) __FUNCTION__ was always documented as a variable, so concatentation of it was always undocumented and so liable to removal at any time
Yes, that's a legally sustainable position, just as it would be sustainable to make an illegal program immediately illegal in the Ada case, but as I said, we have to consider users as well, and telling a user that he cannot sue you for misprepresentation or doing something wrong is not actually helpful to the user trying to get a program running.
The great majority of users do not read all the documentation carefully. They inherit legacy code and learn by example. Yes, if they have to sue you, they will be on weak ground, but you have to be careful that just as you can tell people to RTFM, they can tell you and your compiler to get lost.
So the above kind of argument is weak.
The argument that says: extension X is in practice impossible to maintain without severe damage to efficiency or maintainability is a much stronger argument indeed, you have to balance the pros and cons.
A good example of this is not trying to maintain behaviors for some undefined C situations like overflow and aliasing when optimizing (though for aliasing, we do make a switch avalable, even though we don't have to.
What I think is important is to try to understand and document in advance any incompatibilities, even if they are wrt undocumented features.
