On Saturday, 2 August 2014 at 22:00:27 UTC, Andrew Godfrey wrote:
On Saturday, 2 August 2014 at 21:36:11 UTC, Tobias Pankrath
wrote:
On Saturday, 2 August 2014 at 21:25:40 UTC, Ola Fosheim
Grøstad wrote:
On Saturday, 2 August 2014 at 20:27:09 UTC, Andrei
Alexandrescu wrote:
Hmmm... code that fails assertions is hardly working. --
Andrei
It is not the code that fails the assertion, it is the
asserted proposition that has not be satisfied by the axioms
in the program as it has been formulated in the context. It
does not mean "can not be satisfied", but "has not been
satisfied".
Don't you agree, that a program that throws AssertError in non
-release* build is broken?
* this is not the opposite of debug
By this definition of 'broken', I assert that most shipped
software is broken.
I strongly disagree with that: if there's pressure for a release,
most software in that condition has the assert _removed_ or
_commented_, a bug opened, and a boss directive to do so so that
it's boss responsibility having taken the risk.
The assert is _reinserted_ or _uncommented_ when there's someone
hunting for that bug after the release.
If this operation can't be done because the failing asserts are a
bazillion, compiling it in '-release' with the proposed
assert/assume semantic don't change anything: from time to time,
when the snowball of failed assert-optimisations will start to
destroy the program logic, all that thing will explode in the
face of the user.
---
Paolo
