[Bug 4436] Storage32 regression problem: ECAD app 'Easy-PC' is unable to create new project/design

[Wine Bugs]

http://bugs.winehq.org/show_bug.cgi?id=4436

------- Additional Comments From [e-mail]  2006-09-02 04:58 -------
Thanks for the comments Mike. However I first must stress that the Commit is
_not_ the source of the trouble. The app uses a storage file that is shipped
with the app, i.e. it does not create it. I do have to create one during the
performance tests in order to have something to test with and I do the Commit
just to be sure that all streams are correctly written to the file (I don't know
if it is strictly necessary, as I said I am not a Windows apps programmer!)

The problem is almost certainly somewhere in the reference counting. Yes, my
patch _does_ work on Windows (tested on XP - I can't test other versions of
Win), and _does_ fail on Wine CVS. I described the methodology behind my
Releases in the comments earlier in the bug report.

I completely agree with you that it is technically wrong to have to Release
twice. However, under Wine I _must_ to do this or the storage file is not
closed! This is the actual problem! Under Windows, I only have to Release once,
as would be expected. So that is the reason I had two Releases in the
performance test, the second enclosed in a return code test.

Under Wine, the first Release returns 1, but does not close the storagefile. The
second Release returns zero and closes the file. If I do not execute the second
Release under Wine, the file remains open and cannot be reopened in subsequent
tests.

Under WinXP, the first release returns zero, and closes the storage file. The
second Release is then not executed. This stops the conformance test from
crashing under Windows, as the first Release frees the 'stg' object.

I looked at your patch modification, and this will segfault under Win (I know,
this is how I did it initially!) as after the first Release, the 'stg' object is
freed. The second Release is called on unallocated memory, and the conformance
test crashes. It will, however, not segfault under Wine because under Wine, the
first Release does _not_ free the 'stg' object. The way I did it highlights the
problem without the conformance test crashing under Win.

I have submitted the original patch to wine-patches, however I forgot the
todo_wine { } sections, so I may have to resubmit it.