On 03/04/18 05:12, Huw Davies wrote:

On Fri, Mar 30, 2018 at 12:11:36PM -0500, Sergio Gómez Del Real wrote:
Signed-off-by: Sergio Gómez Del Real <sdelreal@codeweavers.com>
---
 dlls/ole32/datacache.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/dlls/ole32/datacache.c b/dlls/ole32/datacache.c
index 0f23faf08c..452e171e4c 100644
--- a/dlls/ole32/datacache.c
+++ b/dlls/ole32/datacache.c
@@ -1862,6 +1862,7 @@ static HRESULT WINAPI DataCache_Load( IPersistStorage *iface, IStorage *stg )
         DataCacheEntry_Destroy( This, entry );
 
     ReadClassStg( stg, &clsid );
+    This->clsid = CLSID_NULL;
This is probably hiding another bug.  In the new tests you add in
[5/5] the cache entries should have been created when the cache is
created.  Could you investigate what's going on?

Huw.

In create_automatic_entry() we have:
if (IsEqualCLSID( &cache->clsid, clsid )) return S_OK;
which I guess is there to prevent creating 2 static entries of the same CF. When data cache is created with a static picture, its clsid is set with one of CLSID_Picture_*, but only after calling create_automatic_entry(). Then, in _Load, the data cache is flushed (its entries destroyed), but its clsid isn't cleared; it maintains one of CLSID_Picture_* values. The patch does this clearance, which I think makes sense (since we are loading new entries and clsid from a storage), and so when create_automatic_entry() is called right after destroying the entries and reading the storage's clsid, it will succeed.

I would think that it is correct to 'reset' the clsid in _Load, as done in this patch, right after destroying the entries and before reading in the new entries and clsid from the storage; and also that create_automatic_entry() is correct to check the CLSIDs so as to avoid creating a new entry, which wouldn't be the desired behavior. What do you think?

- Sergio