[PATCH] crypt32: Store CRYPT_KEY_PROV_INFO in a platform independent way.

[Dmitry Timoshkov]

Alexandre Julliard <julliard at winehq.org> wrote:

> >> >> >> This makes reading and writing of CRYPT_KEY_PROV_INFO certificate property
> >> >> >> from/to the certificate storage independent from architecture the crypt32.dll
> >> >> >> code being executed on.
> >> >> >
> >> >> > Is there anything that could be improved to make this patch accepted?
> >> >> 
> >> >> That data structure looks strange. Maybe you could explain why it needs
> >> >> to use the 64-bit layout, and why you are achieving that with padding
> >> >> instead of using 64-bit types?
> >> >
> >> > 64-bit layout is used so that 64-bit crypt32 is able to read the structure
> >> > saved by 32-bit crypt32, currently this leads to a memory corruption. There's
> >> > no reason to not use 64-bit types, thanks.
> >> 
> >> That's better, but as long as you are defining your own structure, I
> >> don't see why using simply DWORDs wouldn't work just as well, as long as
> >> you use the same structure on 32- and 64-bit.
> >
> > Obviously DWORDs won't work for pointers, using ULONG64 allows to avoid
> > reallocation when reading CRYPT_KEY_PROV_INFO from the certificate storage,
> > and just fix up the pointers in place as current code is doing.
> 
> I feel it would be cleaner to do that with some sort of union, and use
> the correct types and names for what is being stored. In particular
> keeping names like 'pwszContainerName' for things that are not pointers
> isn't very meaningful. But of course the existing code is already not
> great in that respect...

My main intention was to reuse current code, even it's so much ugly,
that's why I decided to keep using such a structure, and just adapt
it for mixed 32/64-bit case. If you feel strongly about rewriting it,
then current code should be completely replaced.

-- 
Dmitry.