[PATCH 1/2] winmm: Avoid using SuspendThread, it can hang Wine.

Hi,
> this Fixes bug #28388

I think it is not a good idea to defer this patch until after 1.4.0.

Every app using midiStreamOpen eventually calls SuspendThread
which is simply a time bomb on a dual core machine. There are many
more multi-core machines now, so the situation degrades.

Dan Kegel wrote:
> With WINEDEBUG=warn+heap, it happens a lot sooner (first run: first try;
> second run: sixth try).

As noted in the bug report, SuspendThread can hit Wine amid memory
management, breaking havoc. Users won't know who to blame for random
lock-ups and complain like I did that Wine isn't able to generate a backtrace.

Patch 1/2 can be applied stand alone. It gets rid of SuspendThread.

It's been a pleasure to use patch 2/2 and hear apps play midiStream music without trouble.
I wouldn't mind if you defer that for 1.4.1. After all, this is a feature that has been
lacking in Wine for the last 10 years, so people can wait a little more...

Regards,
        Jörg Höhle

[Alexandre Julliard]

Joerg-Cyril.Hoehle(a)t-systems.com writes:
> Hi,
> > this Fixes bug #28388
>
> I think it is not a good idea to defer this patch until after 1.4.0.
>
> Every app using midiStreamOpen eventually calls SuspendThread
> which is simply a time bomb on a dual core machine. There are many
> more multi-core machines now, so the situation degrades.

I don't see any apps in that bug report, it's only about a test failure.
Do you have real apps that trigger the problem in normal usage?

-- 
Alexandre Julliard
julliard(a)winehq.org

> I don't see any apps in that bug report, it's only about a test failure.
> Do you have real apps that trigger the problem in normal usage?

The two apps from #26997 #23756 use midiStream with DCB_CALLBACK_FUNCTION.
They need patch 1+2.

I've not retested ff7 from bug #9220 which uses midiStream too.

The apps that would benefit from patch 1/2 alone are those that
use midiStream with one of DCB_NULL, WINDOW or EVENT.

I've not investigated which MIDI apps are of the required kind or if
they use mcimidi, midiOut directly or midiStream with DCB_FUNCTION.

Regards,
	Jörg Höhle

[Alex Bradbury]

On 17 February 2012 16:14,  <Joerg-Cyril.Hoehle(a)t-systems.com> wrote:
> > I don't see any apps in that bug report, it's only about a test failure.
> > Do you have real apps that trigger the problem in normal usage?
>
> The two apps from #26997 #23756 use midiStream with DCB_CALLBACK_FUNCTION.
> They need patch 1+2.

You mean #26697

Alex