Bug 796 – Asserting a null object reference throws AssertError Failure internal\invariant.d(14) or Access Violation

Status
RESOLVED
Resolution
FIXED
Severity
normal
Priority
P2
Component
dmd
Product
D
Version
D2
Platform
All
OS
All
Creation time
2007-01-05T06:15:00Z
Last change time
2015-06-09T05:11:45Z
Keywords
patch, wrong-code
Assigned to
bugzilla
Creator
lio+bugzilla

Comments

Comment #0 by lio+bugzilla — 2007-01-05T06:15:53Z
The following code causes a run-time assertion failure in internal\invariant.d(14) when linked against a debug build of Phobos (-unittest -g -w): #class Class {} #void main(){ # Class c; # assert(c); #} It causes a run-time Access Violation when linked against the release build of Phobos.
Comment #1 by fvbommel — 2007-01-05T08:48:35Z
It seems to be according to the spec though. I just looked it up, and assert(object) checks object's invariant; not just that it isn't null (that's assert(object !is null) apparently). Not what I would've expected... This isn't exactly clearly noted: this is mentioned on http://www.digitalmars.com/d/class.html#invariants but not http://www.digitalmars.com/d/expression.html#AssertExpression where it would be expected (especially since you wouldn't normally look at the section on invariants when figuring out what an assert does). So maybe this should be noted there too (especially since every other use of objects in boolean context is a null check AFAIK). I've already added it to the comments page for that page. The invariant checking routine in the runtime does perform an assert(o !is null), but that only gets compiled into a non-release build of Phobos. In a release build it instead segfaults trying to look up the vtable of the object. (to get classinfo)
Comment #2 by lio+bugzilla — 2007-01-06T02:54:04Z
Nice catch! I never would have thought of that. In any case, the assertion failure / access violation can be easily fixed if the compiler where to add a null-pointer check prior to checking the invariants. Frankly, I think this makes sense too. We can expect many C/C++ people doing "assert(instance)" only to check the pointer. If it were to check both the pointer and the invariants, that would just make it even more useful. AFAIK, this cannot be changed by adding a check to invariant.d, since that would add unnecessary overhead to all invariant checks. The code generation for assert(instance) should be changed to include a pointer check: the assertion should fail for null-pointers: assert(classref) // => assert( (classref !is null) && _d_invariant(classref) );
Comment #3 by fvbommel — 2007-01-11T04:13:27Z
It seems I misread the spec: it says the invariant *can* be checked on assert(classref), not that it *will* be. So the spec's fine. The bug seems to be in the compiler.
Comment #4 by smjg — 2007-01-11T09:04:37Z
Indeed, assert(classref) should be translated inline to assert((classref !is null) && _d_invariant(classref)); so that the error location is preserved.
Comment #5 by torhu — 2007-04-12T18:55:55Z
I think the assert feature would be more intuitive if assert(obj) would only check for a null reference. That's what people expect it to do, and that's what they use it for most of the time. But doing both kinds of checks would be better than the current behavior. If you really wanted to check the invariant explicitly, the syntax could be 'obj.invariant', 'invariant(obj)' or similar. This could be useful in the class' own methods, where there's no need to check for a null reference anyway.
Comment #6 by andrei — 2010-11-26T14:13:26Z
Still present in 2.050.
Comment #7 by b.helyer — 2011-03-07T03:12:41Z
Every time I write assert(object); A dagger pierces my heart and I remember I must write assert(object !is null); Could we get this fixed this decade some time?
Comment #8 by smjg — 2011-03-07T04:17:53Z
(In reply to comment #7) > Every time I write > > assert(object); > > A dagger pierces my heart and I remember I must write > > assert(object !is null); > > Could we get this fixed this decade some time? Indeed. The intended behaviour (check that object is non-null AND satisfies its invariants) needs to be built into the compiler, not delegated to the RTL.
Comment #9 by viritrilbia+d — 2011-05-31T13:45:01Z
FYI: As a new D programmer, I found the segfault produced by this sort of code very confusing, and spent almost an hour trying to figure out where the issue in my code was and why I wasn't just getting an AssertError. Only when I found this bug report did I have any idea what was going on.
Comment #10 by yebblies — 2011-08-30T10:17:34Z
Comment #11 by alex — 2011-11-08T15:04:55Z
(In reply to comment #10) > https://github.com/D-Programming-Language/dmd/pull/358 +1 to this pull req from me.
Comment #12 by yebblies — 2011-11-08T17:37:10Z
*** Issue 6913 has been marked as a duplicate of this issue. ***
Comment #13 by bugzilla — 2012-01-11T13:30:00Z