TC 11.03RC1 - Lister and video files

[Hacker]

Fla$her,

I don't think the latest fix has reduced the load.

The claim was that the fix reduced the problem (of being annoyed by sound), not that the fix reduced the load.

Roman

[Fla$her]

OK. I read it wrong.

[ghisler(Author)]

The problem with OLE interfaces is that they have a usage counter. When starting to use the interface from somewhere you increase the counter, and when you stop you decrease it. When the counter reaches zero (no one uses the interface any more), the interface gets unloaded.

Now when something else uses the same interface Lister uses but doesn't decrease the counter, the OLE interface stays loaded even when Lister decreases the counter. Lister could continue to decrease the counter until the interface gets unloaded, but then the other user would crash when trying to access (or unload) the now unloaded interface.

[ghisler(Author)]

I have tested this now on an old PC with Windows XP, and I was indeed able to reproduce the problem. It did not occur in the XP emulator.

I did a lot of testing on that old machine, and found out that both stopping the replay and unloading the graph have to be called from the main thread, not from the player thread, although all the other calls are made from the player thread. This only seems to affect Windows XP. The downside is that when stop or unload hang on the main thread, then the entire Total Commander will hang.

So I added the following solution:
1. On Windows XP, call Stop and then unload the graph on the main thread
2. No changes on newer Windows versions
3. If the above change leads to hangs, you can revert to the old method by adding:
wincmd.ini
[Lister]
StopInMainThread=0

I did encounter hangs in Total Commander, but this was because I was still calling "Stop" in the background thread when the track ended.
If this causes Total Commander to hang, I will make StopInMainThread=0 the default.

[Fla$her]

2ghisler(Author)
I'm glad that you have tackled the problem thoroughly. The first thing I thought of when you talked about XP tests was a virtual machine on which it's not always possible to detect a problem. We've seen this before with other bugs.

In RC5 threads began to unload (except for new ones from RPCRT4.dll), no pause is required, no freezes detected!
If it's not possible to unload the calm thread with RPCRT4.dll, then in any case, I consider the problem solved for myself, i.e. the topic can be moved here.

[ghisler(Author)]

Nice to hear that! I'm not seing any threads from RPCRT4.dll, and after closing the first video, the number of threads does not increase each time I close another video. Do you get one more thread after each video is closed?

[Fla$her]

I withdraw the question. I checked without plugins — everything is fine. You can safely move the topic.

[ghisler(Author)]

Great, thanks!

Moderator message

Moved to fixed bugs