Re: [13/13] fusion: Add tests for the IAssemblyEnum interface.

[Paul Vriens]

Hans Leidekker wrote:
> -Hans

Hi Hans,

These tests crash on Vista with .NET 3.5 (not sure about other platforms/.NET 
versions yet). The enum_gac_assemblies function relies on the fact that 
ffd.cFileName contains 2 underscores. On Vista however I see some assemblies 
with a different format for the cFileName like:

6.0.0.0_en_31bf3856ad364e35

This explains the (first) crash as we dereference a pointer. So doing something 
like this fixes that piece:

else if (depth == 1)
          {
-            ptr = strstr(ffd.cFileName, "__");
+            CHAR version[MAX_PATH];
+
+            lstrcpyA(version, ffd.cFileName);
+            ptr = strstr(version, "_");
              *ptr = '\0';
-            ptr += 2;
+            ptr = strrchr(ffd.cFileName, '_');
+            ptr += 1;
              sprintf(buf, "Version=%s, Culture=neutral, PublicKeyToken=%s",
-                    ffd.cFileName, ptr);
+                    version, ptr);

This seems to work fine but than it crashes again while trying to free the 
memory "HeapFree(GetProcessHeap(), 0, asmname->data);" in test_enumerate.

Just ignoring those HeapFree's doesn't fix it as we crash again later on.

[13/13] fusion: Add tests for the IAssemblyEnum interface.

[Hans Leidekker]

Hi Paul,
> These tests crash on Vista with .NET 3.5 (not sure about other platforms/.NET 
> versions yet). The enum_gac_assemblies function relies on the fact that 
> ffd.cFileName contains 2 underscores. On Vista however I see some assemblies 
> with a different format for the cFileName like:
>
> 6.0.0.0_en_31bf3856ad364e35

Yes, that test simply ignores the culture attribute. Is that one of your
own machines? If so, can you try this patch?

-Hans

[13/13] fusion: Add tests for the IAssemblyEnum interface.

[Paul Vriens]

Hans Leidekker wrote:
> Hi Paul,
> > These tests crash on Vista with .NET 3.5 (not sure about other platforms/.NET 
> > versions yet). The enum_gac_assemblies function relies on the fact that 
> > ffd.cFileName contains 2 underscores. On Vista however I see some assemblies 
> > with a different format for the cFileName like:
> >
> > 6.0.0.0_en_31bf3856ad364e35
>
> Yes, that test simply ignores the culture attribute. Is that one of your
> own machines? If so, can you try this patch?
>
> -Hans

At least we don't crash at that point (but still at the HeapFree). The patch is 
not correct however as tracing shows in some cases:

Culture=en_31bf3856ad364e35

(and that makes sense reading the patch ;) )

[13/13] fusion: Add tests for the IAssemblyEnum interface.

[Paul Vriens]

Paul Vriens wrote:
> Hans Leidekker wrote:
> > Hi Paul,
> > > These tests crash on Vista with .NET 3.5 (not sure about other 
> > > platforms/.NET versions yet). The enum_gac_assemblies function relies 
> > > on the fact that ffd.cFileName contains 2 underscores. On Vista 
> > > however I see some assemblies with a different format for the 
> > > cFileName like:
> > >
> > > 6.0.0.0_en_31bf3856ad364e35
> >
> > Yes, that test simply ignores the culture attribute. Is that one of your
> > own machines? If so, can you try this patch?
> >
> > -Hans
>
> At least we don't crash at that point (but still at the HeapFree). The 
> patch is not correct however as tracing shows in some cases:
>
> Culture=en_31bf3856ad364e35
>
> (and that makes sense reading the patch ;) )

The second crash (HeapFree) is because we use strdup. Looks like on Windows 
strdup doesn't allocate memory on heap.

I guess we should replace strdup by a HeapAlloc/lstrcpyA ?