https://issues.dlang.org/show_bug.cgi?id=14336

Dmitry Olshansky <dmitry.olsh@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|major                       |critical

--

https://issues.dlang.org/show_bug.cgi?id=14336

Dmitry Olshansky <dmitry.olsh@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|nobody@puremagic.com        |dmitry.olsh@gmail.com

--

https://issues.dlang.org/show_bug.cgi?id=14336

--- Comment #1 from Dmitry Olshansky <dmitry.olsh@gmail.com> ---
(In reply to Vladimir Panteleev from comment #0)
> Created attachment 1495 [details]
> Valgrind log
> 
> Valgrind (with my Valgrind branch [1]) reports an incorrect memory access in
> std.uni.CowArray destructors (see attachment).
> 
> It looks like the following happens:
> 
> - The destructor is finalizing a heap-allocated array of
> InversionList!GcPolicy structs.
> - InversionList doesn't have a destructor, but it has a CowArray field
> ("data"). CowArray has a destructor, so one is automatically generated for
> InversionList.
> - CowArray!GcPolicy.~this calls the refCount @property.
> - The refCount @property attempts to refer to the heap-allocated (via
> GcPolicy) uint[] data field, which has already been destroyed by the GC. As
> I understand, this is an invalid memory access.
> 

Do you know a way to see if we are in GC finalizer vs normal destructor?

--

https://issues.dlang.org/show_bug.cgi?id=14336

safety0ff.bugz <safety0ff.bugz@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |safety0ff.bugz@gmail.com

--- Comment #2 from safety0ff.bugz <safety0ff.bugz@gmail.com> ---
(In reply to Dmitry Olshansky from comment #1)
> 
> Do you know a way to see if we are in GC finalizer vs normal destructor?

gc_inFinalizer discussion here: https://issues.dlang.org/show_bug.cgi?id=17563

--

https://issues.dlang.org/show_bug.cgi?id=14336

Carsten Blüggel <chilli@posteo.net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |chilli@posteo.net

--- Comment #3 from Carsten Blüggel <chilli@posteo.net> ---
Maybe the following PR is related (or close to) this issue?
It jail's previously possible (as per dmd analysis) escaping this.data pointer
through InversionList.byInterval() with -dip1000.
https://github.com/dlang/phobos/pull/6041

--

https://issues.dlang.org/show_bug.cgi?id=14336

--- Comment #4 from Carsten Blüggel <chilli@posteo.net> ---
(In reply to Vladimir Panteleev from comment #0)
> Valgrind (with my Valgrind branch [1]) reports an incorrect memory access in
> std.uni.CowArray destructors (see attachment).

I submitted https://github.com/dlang/phobos/pull/6041 to fix a std.uni.d
-dip1000 issue (there is no @trusted in InversionList!GcPolicy, CowArray or
GcPolicy any more (only ReallocPolicy currently still has @trusted), thus
theoretically it should be memory safe for GcPolicy) and want to check, whether
Your issue is fixed as well.
I assume, current druntime code doesn't cover memcheck, thus Your valgrind
branch is still required. Due to my current lack of sufficient druntime
knowledge, I failed to merge Your valgrind branch (conflicts in
src/gc/impl/conservative/gc.d).

Will You please update Your valgrind branch or at least conservative/gc.d to be mergable with current druntime/master, or give me a hint how to do that?

Thanks :-)

--

https://issues.dlang.org/show_bug.cgi?id=14336

--- Comment #5 from Dmitry Olshansky <dmitry.olsh@gmail.com> ---
(In reply to Carsten Blüggel from comment #4)
> (In reply to Vladimir Panteleev from comment #0)
> > Valgrind (with my Valgrind branch [1]) reports an incorrect memory access in
> > std.uni.CowArray destructors (see attachment).
> 
> I submitted https://github.com/dlang/phobos/pull/6041 to fix a std.uni.d
> -dip1000 issue (there is no @trusted in InversionList!GcPolicy, CowArray or
> GcPolicy any more (only ReallocPolicy currently still has @trusted), thus
> theoretically it should be memory safe for GcPolicy) and want to check,
> whether Your issue is fixed as well.
> I assume, current druntime code doesn't cover memcheck, thus Your valgrind
> branch is still required. Due to my current lack of sufficient druntime
> knowledge, I failed to merge Your valgrind branch (conflicts in
> src/gc/impl/conservative/gc.d).
> 


Sadly it won't fix low-level detail of using the data allocated from GC during finalization.

On the other hand that makes Finalizers in GC pretty much useless as you can't touch any memory.

--

https://issues.dlang.org/show_bug.cgi?id=14336

Walter Bright <bugzilla@digitalmars.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |safe
                 CC|                            |bugzilla@digitalmars.com

--

https://issues.dlang.org/show_bug.cgi?id=14336

Iain Buclaw <ibuclaw@gdcproject.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P1                          |P2

--