Re: [v2 PATCH 1/3] ucrtbase: Implement _initialize_onexit_table()

[Piotr Caban]

On 04/24/16 20:51, Nikolay Sivov wrote:
> dlls/ucrtbase/tests/onexit.c                       | 114 +++++++++++++++++++++
> Probably we don't need a separate file just for these tests. Could you 
> please put them into a file with more generic name so it can be used for 
> other tests in future.
> +    memset(&table, 0, sizeof(table));
> +    ret = p_initialize_onexit_table(&table);
> +    ok(ret == 0, "got %d\n", ret);
> +    ok(table._first == table._last && table._first == table._end, "got first %p, last %p, end %p\n",
> +        table._first, table._last, table._end);
> +todo_wine
> +    ok(table._first != NULL, "got %p\n", table._first);
> We probably don't need to be binary compatible but native uses following 
> algorithm (x - random number, 32-bit version):
> void* encode_ptr(void *p) {
>      DWORD r = (DWORD)p;
>      r = _rotl(r, x);
>      return r^x;
> }
> void* decode_ptr(void *p) {
>      DWORD r = (DWORD)p;
>      r = r^x;
>      return _rotr(r, x);
> }
> In theory this test may fail if x happens to be 0.

Thanks,
Piotr

[v2 PATCH 1/3] ucrtbase: Implement _initialize_onexit_table()

[Nikolay Sivov]

On 25.04.2016 18:54, Piotr Caban wrote:
> On 04/24/16 20:51, Nikolay Sivov wrote:
> > dlls/ucrtbase/tests/onexit.c                       | 114
> > +++++++++++++++++++++
> > Probably we don't need a separate file just for these tests. Could you
> > please put them into a file with more generic name so it can be used for
> > other tests in future.

Sure.
> > +    memset(&table, 0, sizeof(table));
> > +    ret = p_initialize_onexit_table(&table);
> > +    ok(ret == 0, "got %d\n", ret);
> > +    ok(table._first == table._last && table._first == table._end,
> > "got first %p, last %p, end %p\n",
> > +        table._first, table._last, table._end);
> > +todo_wine
> > +    ok(table._first != NULL, "got %p\n", table._first);
> > We probably don't need to be binary compatible but native uses following
> > algorithm (x - random number, 32-bit version):
> > void* encode_ptr(void *p) {
> >     DWORD r = (DWORD)p;
> >     r = _rotl(r, x);
> >     return r^x;
> > }
> > void* decode_ptr(void *p) {
> >     DWORD r = (DWORD)p;
> >     r = r^x;
> >     return _rotr(r, x);
> > }
> > In theory this test may fail if x happens to be 0.

Yes, it's similar to Encode/DecodePointer(), with different parameters,
likely internal to CRT. I don't think we need to be that compatible, at
least I don't have any evidence right now that we should.
> Thanks,
> Piotr