[Bug 34343] New: mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

Bug #: 34343
           Summary: mIRC 7.32: entering infinite loop after dismissing
                    trial dialog on Mac OS X
           Product: Wine
           Version: 1.7.0
          Platform: x86
        OS/Version: Mac OS X
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: -unknown
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
    Classification: Unclassified

Created attachment 45716
  --> http://bugs.winehq.org/attachment.cgi?id=45716
Excerpt from WINEDEBUG=warn+all,err+all,fixme-all,trace+relay wine mirc.exe

A newly installed mIRC in a clean wineprefix is entering a closed loop after
dismissing the trial dialog on Mac OS X.
It doesn't matter where it was installed, I also tried dosdrive_c/mIRC, just to
make sure.

Excerpt from WINEDEBUG=warn+all,err+all,fixme-all,trace+relay wine mirc.exe is
attached.

I can upload the full log, if need be, or logs with other WINEDEBUG values
(though trace tends to be very verbose.)

This bug is only affecting OS X, mIRC is starting up and working fine (mostly)
with wine 1.7.0 on Linux. Don't try to reproduce it on Linux, please.

Wine has been built and installed via MacPorts.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

Mihai Moldovan  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |download
                 CC|                            |[e-mail]

--- Comment #1 from Mihai Moldovan  2013-08-24 07:57:30 CDT ---
Download URL: http://www.mirc.co.uk/get.html

Also adding download keyword.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

Ken Thomases  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #2 from Ken Thomases  2013-08-24 16:38:56 CDT ---
I'm not able to reproduce this using either Wine 1.7.0 or current git. I
downloaded mIRC 7.32 from the link you gave. I installed it into a new
wineprefix. I ran the app. I got the dialog telling me I had 30 days to
evaluate and letting me Register or Continue; I chose Continue. I was then able
to log into FreeNode and join #winehackers.

What version of Mac OS X are you using? I'm running 10.6.8.

Are any wine processes using lots of CPU? Can you collect and attach a sample
report from those? (You can either do View > Sample Process in Activity Monitor
or use the command "sample <pid>" in Terminal.)

If none are using lots of CPU, see if you can identify the PID of the one
running mIRC and sample that. (Check the output of "ps xww | grep wine" to find
the right process.)

Does it make any difference if you switch to using the X11 driver instead of
the Mac driver? Use regedit to set this registry key:

[HKEY_CURRENT_USER\Software\Wine\Drivers]
"Graphics"="x11"

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #3 from Mihai Moldovan  2013-08-24 17:52:04 CDT ---
(In reply to comment #2)
> I'm not able to reproduce this using either Wine 1.7.0 or current git.

Bummer. :(
> What version of Mac OS X are you using? I'm running 10.6.8.

OS X 10.8.4 (Snow Leopard) here.
> Are any wine processes using lots of CPU? Can you collect and attach a sample
> report from those? (You can either do View > Sample Process in Activity Monitor
> or use the command "sample <pid>" in Terminal.)

Yes, of course. The mIRC process does. :)
No other wine processes, though.
> If none are using lots of CPU [...]

Sample file incoming.
> Does it make any difference if you switch to using the X11 driver instead of
> the Mac driver?

Unfortunately, no, the behavior is exactly the same. Thank you for the handy
trick for switching between quartz and X11, though!

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #4 from Mihai Moldovan  2013-08-24 17:54:25 CDT ---
Created attachment 45719
  --> http://bugs.winehq.org/attachment.cgi?id=45719
Sample file auf wine mirc.exe process.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

Mihai Moldovan  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #45719|Sample file auf wine        |Sample file of wine
        description|mirc.exe process.           |mirc.exe process.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #5 from Mihai Moldovan  2013-08-24 18:09:22 CDT ---
(In reply to comment #3)
> OS X 10.8.4 (Snow Leopard) here.

Sorry, Mountain Lion... what is it with those cats.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #6 from Ken Thomases  2013-08-24 18:32:08 CDT ---
Created attachment 45720
  --> http://bugs.winehq.org/attachment.cgi?id=45720
Disable TLS 1.1+ on Mac

Hmm.  It definitely seems to be a bug.  It looks like
dlls/user32/button.c:BUTTON_CheckAutoRadioButton() is looping infinitely
because dlls/user32/dialog.c:GetNextDlgGroupItem() is never returning the
starting control after the first time.

I'm not sure why that happens for you and not me.  One difference that comes to
mind between 10.6.x and 10.8.x is the support for more recent versions of TLS
in secur32.  I've attached a patch which neutralizes that difference.  It
doesn't address the actual bug in user32, but maybe changes mIRC's behavior to
avoid it.  (Although, I would expect that TLS 1.1+ would be supported on most
Linux distros, so it doesn't explain why it works there.)

Does it help?

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #7 from Mihai Moldovan  2013-08-24 18:44:32 CDT ---
(In reply to comment #6)
> I'm not sure why that happens for you and not me.  One difference that comes to
> mind between 10.6.x and 10.8.x is the support for more recent versions of TLS
> in secur32.

Why would support a newer TLS version ever cause problems like this?

I'll give it a try though...

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #8 from Ken Thomases  2013-08-24 18:47:35 CDT ---
Well, it's nothing more than a hunch, really.  My thinking is that mIRC will
set different options in its Connect dialog (the one that should show up after
continuing from the trial dialog) depending on what TLS versions are supported
by the platform.  I further figured, it is during this option-setting that the
infinite loop bug is triggered.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #9 from Mihai Moldovan  2013-08-24 18:53:15 CDT ---
Ah, hmm... that kinda makes sense. Recompiling wine takes ages though, sorry.

Please forgive my inability to type correctly... run a mental s/support
a/support for a/ on the last comment.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #10 from Mihai Moldovan  2013-08-24 19:35:38 CDT ---
Created attachment 45722
  --> http://bugs.winehq.org/attachment.cgi?id=45722
Sample with TLS 1.1 disabled.

Sorry for the previous sample, as well, because that was created with wine 1.6.
But as it's showing the same symptoms as 1.7.0, I guess we can ignore it.

Still no luck, even with disabling TLS 1.1+. Looping and looping.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #11 from Mihai Moldovan  2013-08-25 05:55:17 CDT ---
One other difference comes to mind, though. You're likely building wine with
(llvm-?)gcc, while MacPorts defaults to clang on OS X 10.8. Same thing for
Linux, which has been using GCC to build wine. I guess I could/should try
rebuilding with gcc.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #12 from Bruno Jesus  2013-08-25 07:57:01 CDT ---
Reminds me of http://bugs.winehq.org/show_bug.cgi?id=34066#c3

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #13 from Mihai Moldovan  2013-08-25 08:07:32 CDT ---
(In reply to comment #12)
> Reminds me of http://bugs.winehq.org/show_bug.cgi?id=34066#c3

I've seen similar errors with .NET installers. Likewise built with clang. It
looped, but I didn't really care to debug it. May be the very same symptom,
though, as looping started after clicking a button.

I'm about to install Apple's GCC 4.2, because... Apple removed "pure" GCC as of
10.7 and we're now stuck with llvm-gcc, which doesn't compile wine correctly
(crashing on startup.)

Likewise, I have FSF GCCs 4.4 to 4.7 installed on my system, but those won't
compile the winemac.drv due to missing ObjC blocks support.

So fetching and installing Apple GCC 4.2 it is, which will likely take some
time.

Unrelated to this, I'll also try building wine with clang-3.3 and see if that
behaves better.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #14 from Mihai Moldovan  2013-08-25 11:13:03 CDT ---
Wine git compiled with clang 3.3 on Linux seems to behave as if compiled with
GCC, no loops (weirdly?)

Moving on to compiling wine with clang 3.3 on OS X and building Apple GCC
4.2.1. This will probably take a good while...

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #15 from Mihai Moldovan  2013-08-25 12:36:42 CDT ---
(In reply to comment #14)
> Wine git compiled with clang 3.3 on Linux seems to behave as if compiled with
> GCC, no loops (weirdly?)
>
> Moving on to compiling wine with clang 3.3 on OS X...

Aaaaand that's working. No more loop, the connection window comes right up
after dismissing the trial dialog. Does this mean, that the bug is likely being
triggered by clang 3.2 as provided by Apple?!
I'm also having a vanilla clang 3.2 compiled with MacPorts, guess I'll try
reproducing the bug with that one too. Not that I wouldn't trust Apple... screw
that, I very much don't.
> and building Apple GCC 4.2.1.

Still working/compiling on that.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #16 from Ken Thomases  2013-08-25 12:44:58 CDT ---
I'm guessing it has to do with the compiler's attempts to optimize the
generated code.  If you have a build that reproduces the problem and then you
do:

cd dlls/user32
touch button.c dialog.c
make CFLAGS=-O0

and then run that Wine from within the build tree, I suspect that will avoid
the problem.

There are a couple of possibilities: it's a bug in the compiler such that it's
producing incorrect code from the Wine sources, or it's a subtle bug in Wine
that's only revealed by certain optimization strategies.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #17 from Mihai Moldovan  2013-08-25 13:51:11 CDT ---
(In reply to comment #16)
> I'm guessing it has to do with the compiler's attempts to optimize the
> generated code.

This said, I have actually not passed any optimization flags when compiling
wine git with clang 3.3 (neither on OS X, nor on Linux.) I figured
configure/the Makefiles would default to -O2. Instead, there's no optimization
taking place (as I can see in the current compilation output of both apple gcc
4.2 and vanilla clang 3.2.)

Thanks for catching that, I'll take a look at what the Linux box does and will
also play around with optimization levels.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #18 from Mihai Moldovan  2013-08-25 14:57:00 CDT ---
Wow, I'm speechless.

Here's a first result list:

wine 1.6 + Apple clang 3.2 + opt (compiled by MacPorts): failing
wine 1.7.0 + Apple clang 3.2 + opt (compiled by MacPorts): failing

wine git (8f09d3470) + Apple GCC 4.2 + noopt: working
wine git (8f09d3470) + vanilla clang 3.3 + noopt: working
wine git (8f09d3470) + vanilla clang 3.2 + noopt: working
wine git (8f09d3470) + Apple clang 3.2 + noopt: failing

All those wine versions have been built using exactly the same configure
arguments and CFLAGS, the only alternation has been the CC environment
variable.

---

Linux:
wine git (8f09d3470) + (Gentoo) GCC 4.7 + opt: working
wine git (8f09d3470) + (Gentoo) clang 3.3 + noopt: working

---

What is this? Could this actually be a compiler bug, because Apple decided to
ship a pre-release of clang 3.2 which contains bugs, as opposed to the released
clang 3.2 version?

That's what /usr/bin/clang --version tells me:
Apple LLVM version 4.2 (clang-425.0.28) (based on LLVM 3.2svn)

I guess I'm still gonna rebuild wine with all those compilers for fun with
optimizations enabled and see if that behaves any differently.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #19 from Mihai Moldovan  2013-08-25 18:28:46 CDT ---
Results with optimization (-O2):

wine git (8f09d3470) + vanilla clang 3.3 + opt: working
wine git (8f09d3470) + vanilla clang 3.2 + opt: working
wine git (8f09d3470) + Apple GCC 4.2 + opt: working
wine git (8f09d3470) + Apple clang 3.2 + opt: failing

Does this mean it's a compiler bug and we can't do anything on wine side, but
rather make MacPorts blacklist Apple clang 3.2 for wine and compile using the
MacPorts provided clang 3.2 or 3.3 ports? :(

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #20 from Ken Thomases  2013-08-25 18:53:18 CDT ---
(In reply to comment #19)
> Does this mean it's a compiler bug and we can't do anything on wine side, but
> rather make MacPorts blacklist Apple clang 3.2 for wine and compile using the
> MacPorts provided clang 3.2 or 3.3 ports? :(

Sounds like it.  Thanks for doing all of that building and testing.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

http://bugs.winehq.org/show_bug.cgi?id=34343

--- Comment #21 from Austin English  2013-08-26 14:37:00 CDT ---
This is http://llvm.org/bugs/show_bug.cgi?id=9707, it's fixed in Clang's trunk.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

https://bugs.winehq.org/show_bug.cgi?id=34343

Mihai Moldovan  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |UPSTREAM

--- Comment #22 from Mihai Moldovan  ---
Fixed in clang upstream.

[Bug 34343] mIRC 7.32: entering infinite loop after dismissing trial dialog on Mac OS X

https://bugs.winehq.org/show_bug.cgi?id=34343

Bruno Jesus  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #23 from Bruno Jesus  ---
(In reply to Mihai Moldovan from comment #22)
> Fixed in clang upstream.

Closing upstream.