[PATCH 3/3] shell32/autocomplete: Revamp pwzsRegKeyPath handling so it can deal with arbitrary sizes and make it more robust

[Gabriel Ivăncescu]

On Thu, Aug 30, 2018 at 2:50 PM, Alexandre Julliard <julliard at winehq.org> wrote:
> Gabriel Ivăncescu <gabrielopcode at gmail.com> writes:
>
> There can still be a race, only you are making the function fail in that
> case, that's not nice behavior. The right way to do that sort of thing
> is to allocate a reasonable buffer, get the string (without querying the
> length first), and on overflow restart the loop with a larger buffer.
>

Okay, you are right. I'll have it get the length first (because this
simplifies the case where the value doesn't exist in HKCU but it does
in HKLM), and start with that buffer size, then use a loop until it
returns success (or some error other than ERROR_MORE_DATA) or its type
is not REG_SZ (this should be a failure, right?).

>
> The Wine implementation replicates the Windows behavior, based on test
> cases. That's more reliable than MSDN.
>

Yes you are correct but some hooks (like madCodeHook used in jauntePE)
might hook the registry APIs to provide virtualized redirections. They
*might* follow the MSDN instead of test-cases, because that's what
MSDN claims, I really have no idea if they'll return a non-NUL
terminated string though. Was just playing it safe. At the very least,
should I just forcefully NUL terminate the buffer? (without increasing
its size by +1) It would only be a single extra store and doesn't do
anything on an already NUL terminated buffer.