On Saturday, 2 August 2014 at 05:59:14 UTC, Timon Gehr wrote:
On 08/02/2014 05:34 AM, Andrew Godfrey wrote:
Suppose I call some logging function which has a faulty
assertion in it.
What about Walter's position prevents that assertion's effects
from
escaping the logging function and infecting my code?
Nothing. Undefined behaviour is completely non-modular.
I know cross-module
optimization is hard hence this may be unlikely,
Inlining may happen.
but still it shows
something missing.
I know Walter has said repeatedly that at this point the
program is
invalid, Implying that it deserves whatever happens to it. I
suspect
this is a subtopic we should be discussing more directly. What
are the
boundaries?
The semantics that are proposed to be pulled into the language
spec are currently summarized as:
- In non-release mode, an assertion failure will abort the
program (maybe by throwing an Error).
- In release mode, an assertion failure leads to undefined
behaviour.
I.e. I think there aren't any 'boundaries', unless I am
misunderstanding this term.
So even if the assertion is incorrect and the code is correct,
the caller's correctness can be compromised? This seems
impractical. As in, no one will in good conscience turn it on,
except when their program is very small.
