[Bug 50625] New: Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

Bug ID: 50625
           Summary: Unhandled exception with try/catch
           Product: Wine
           Version: 4.12.1
          Hardware: arm
                OS: Linux
            Status: UNCONFIRMED
          Severity: critical
          Priority: P2
         Component: ntdll
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

Created attachment 69319
  --> https://bugs.winehq.org/attachment.cgi?id=69319
Test Project File

Hello,

I  have run into a bug with try/catch blocks throwing an unhandled exception
even though I have a catch all.

I  have attached a very simple test program that when started calls a method
that just throws an exception class back to the caller.  This very simple
program works great in Windows and throws an unhandled exception in Wine and
terminates the program.

This is a critical path to my application as I have incorporated a 3rd party
library that uses exceptions for program flow rather than function return
codes. Without a fix our application becomes unstable and crashes when an
intended exception is thrown.

I would really like a solution for the version of Wine that we are running
(4.12.1) as we have Wine running on thousands of remote computers with not easy
solution to upgrade to newer versions.  I can however deploy patches to Wine.

Thank you for any assistance with addressing this issue as I have spent the
last couple days annotating the Wine sources to try and find the root cause
without success to this point.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

Austin English  changed:

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

--- Comment #1 from Austin English  ---
Hi David,

Given that you're testing on a (relatively) untested platform (arm) _and_ an
old version, have you verified that this still occurs in current (6.x) wine
(your bug report doesn't specify).

There's obviously no work ongoing on the 4.x branch, but if still present in
6.x, it's worth fixing (and maybe backporting).

Anyway, I can't test this at the moment (maybe this weekend), but CC'ing
myself/Andre, as I know he's interested in wine/arm.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #2 from David Paradis  ---
Austin,

Thank you for looking into this.  Unfortunately I do not have the easy option
of trying this with the latest version.  As I put in the description we have
several hundred if not more of devices installed around the US that are running
4.12.1 and it is not a simple task to upgrade those devices.  We run a building
automation system with a piece of hardware running in a remote store.  I would
like to at some point in the future look at what it will take for us to use the
latest, but that is not an option at the moment.

I report a similar but back in Dec for issues with std::mutex that I was able
to create my own fix.  The bug was closed after fixing it in the upcoming 6.2
release.  I was ok with that because I basically put the same fix in myself. 
With this issue I have not be able to find the root cause to create my own fix.

If you are able to test this in ARM with the 4.12.1 release and let me know
what/where to make the fix I have no problem putting that into my code base and
testing.

Again, thank you very much for whatever help you are able to give.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

Austin English  changed:

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

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #3 from Austin English  ---
Sure, I understand that you can't test on all the devices in field, but
particularly for a relatively exotic platform (arm), I figured you'd have a
test box around that you could try on ;)

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #4 from David Paradis  ---
Understood.

As you know with ARM there are no pre-built releases.  The last time that I had
to configure and compile the sources it took nearly 20hrs to build them.

I do have a test box that is my development machine.  Trying to get the latest
up and running on it would take me the entire weekend to do.  If need be I can
possibly look into do that next week.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #5 from Alexandre Julliard  ---
Exception unwinding in native ARM binaries is not implemented, so I'm afraid
this is not going to be a simple fix.

It would be helpful if you could attach a compiled binary of the test case.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #6 from David Paradis  ---
Created attachment 69323
  --> https://bugs.winehq.org/attachment.cgi?id=69323
Compiled test application

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #7 from David Paradis  ---
That is interesting because I have put in place some FIXME messages to see the
code path and it gets to the point of raising the exception and then crashes. 
I would have assumed if it wasn't implemented that there wouldn't be any
implementation code in files specifically named /ntdll/signal_arm.c and
arm64.c.  We are working with a 32bit arm processor.

Do you know whether it has been implemented in the latest versions?

I have attached a compiled 32bit arm test application.

Again thank you for the assistance on this, I really do appreciate any help you
can give as this is a critical piece of new functionality that we are in the
beginning phase of development on.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #8 from André H.  ---
Please try again with wine 6.22, it's very likely fixed

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #9 from David Paradis  ---
(In reply to André H. from comment #8)
> Please try again with wine 6.22, it's very likely fixed

Thank you Andre.  I will download and and try this release and make sure that
it is has been addressed.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #10 from David Paradis  ---
(In reply to David Paradis from comment #9)
> (In reply to André H. from comment #8)
> > Please try again with wine 6.22, it's very likely fixed
>
> Thank you Andre.  I will download and and try this release and make sure
> that it is has been addressed.

Hello Andre,

I have built and am trying to test the SEH unwinding and having some issues.

I am getting the following messages (and lots of them) when my application
calls "throw ExceptionClass();".
010c:err:seh:unwind_builtin_dll libunwind not available, unable to unwind

I have done an "apt install libunwind-dev" thinking that, that would fix the
issue, but still getting the same result.

This is the command that I am using the configure the Wine sources before doing
a make:
./configure --with-xml --with-png --with-netapi --with-gnutls --without-x
--without-freetype --without-coreaudio --without-mpg123 --without-pulse
--without-sane --without-vdk3d

Am I missing are argument to including the unwind library?  I tried to add
"--with-libunwind" but that argument was unrecognized.

Again, any help is much appreciated.

Thank you,
David Paradis

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

Zebediah Figura  changed:

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

--- Comment #11 from Zebediah Figura  ---
The configure flag you're looking for is "--with-unwind", which can be found by
running "configure --help".

I suspect you need the multiarch version of libunwind-dev. I'm not sure of the
correct code, but I think you want to install libunwind-dev:armhf.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #12 from David Paradis  ---
(In reply to Zebediah Figura from comment #11)
> The configure flag you're looking for is "--with-unwind", which can be found
> by running "configure --help".
>
> I suspect you need the multiarch version of libunwind-dev. I'm not sure of
> the correct code, but I think you want to install libunwind-dev:armhf.

Thank you.  I will try that right now.

When I ran the install for unwind it did download the armf.deb versions, not
the hf.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #13 from David Paradis  ---
(In reply to Zebediah Figura from comment #11)
> The configure flag you're looking for is "--with-unwind", which can be found
> by running "configure --help".
>
> I suspect you need the multiarch version of libunwind-dev. I'm not sure of
> the correct code, but I think you want to install libunwind-dev:armhf.

No change in behavior.  When I ran ./configure -with-unwind I saw it come
across the screen when it was looking for the unwind files to enable it, it
reported "no".  When I then compile the sources I can see in the
dlls/ntdll/linux/signal-arm.c in the unwind_builtin_dll() that the #define for
HAVE_LIBUNWIND is still not being defined, if I attempt to force it to be
defined it compiles but will not link because the lib is not included on the
linker portion of the build.

I see in /usr/include that the libunwind.h file is there and in the
/usr/lib/arm-linux-gnueabihf folder that there are several libunwind*.so files
present.  Not sure why the .configure is disabling the unwind library.

Do you have anything that you could suggest to get this to include it, as
everything looks correct?

I am attaching the output of the .configure command as well as the configure
log that was auto generated, hopefully those help to understand what is going
on.

Again, thank you.
David

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #14 from David Paradis  ---
Created attachment 71180
  --> https://bugs.winehq.org/attachment.cgi?id=71180
.Configure auto log

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #15 from David Paradis  ---
Created attachment 71181
  --> https://bugs.winehq.org/attachment.cgi?id=71181
.Configure output redirect

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

André H.  changed:

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

--- Comment #16 from André H.  ---
CCing Martin
https://stackoverflow.com/questions/49438381/getting-linker-errors-when-link...
looks similar

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #17 from David Paradis  ---
(In reply to André H. from comment #16)
> CCing Martin
> https://stackoverflow.com/questions/49438381/getting-linker-errors-when-
> linking-against-libunwind-for-a-library-in-my-project looks similar

Yes it sounds similar, but when you look at the signal_arm.c file, I attempted
to just force the "HAVE_LIBUNWIND" define just before it looked for it to be
defined.  Inside the ifdef it defines the UNW_LOCAL_ONLY and then includes the
header.

I am trying overload the UNWIND_LIBS path when calling .configure to see if
that helps.

When I then run make I get the following output:

admin@eSCi-EDP001:~/wine-6.22/dlls/ntdll$ sudo make prefix=/opt/wine install
make[1]: Entering directory '/home/admin/wine-6.22'
gcc -o dlls/ntdll/ntdll.so -shared -Wl,-Bsymbolic -Wl,-soname,ntdll.so
-Wl,-z,defs dlls/ntdll/unix/cdrom.o \
  dlls/ntdll/unix/debug.o dlls/ntdll/unix/env.o dlls/ntdll/unix/file.o
dlls/ntdll/unix/loader.o \
  dlls/ntdll/unix/loadorder.o dlls/ntdll/unix/process.o
dlls/ntdll/unix/registry.o \
  dlls/ntdll/unix/security.o dlls/ntdll/unix/serial.o dlls/ntdll/unix/server.o
\
  dlls/ntdll/unix/signal_arm.o dlls/ntdll/unix/signal_arm64.o
dlls/ntdll/unix/signal_i386.o \
  dlls/ntdll/unix/signal_x86_64.o dlls/ntdll/unix/socket.o
dlls/ntdll/unix/sync.o \
  dlls/ntdll/unix/system.o dlls/ntdll/unix/tape.o dlls/ntdll/unix/thread.o
dlls/ntdll/unix/virtual.o \
  dlls/ntdll/unix/version.o -lpthread -ldl
dlls/ntdll/unix/signal_arm.o: In function `unwind_builtin_dll':
/home/admin/wine-6.22/dlls/ntdll/unix/signal_arm.c:246: undefined reference to
`_ULarm_init_local'
/home/admin/wine-6.22/dlls/ntdll/unix/signal_arm.c:253: undefined reference to
`_ULarm_get_proc_info'
/home/admin/wine-6.22/dlls/ntdll/unix/signal_arm.c:277: undefined reference to
`_ULarm_step'
/home/admin/wine-6.22/dlls/ntdll/unix/signal_arm.c:289: undefined reference to
`_ULarm_get_reg'
/home/admin/wine-6.22/dlls/ntdll/unix/signal_arm.c:290: undefined reference to
`_ULarm_get_reg'
/home/admin/wine-6.22/dlls/ntdll/unix/signal_arm.c:291: undefined reference to
`_ULarm_get_reg'
/home/admin/wine-6.22/dlls/ntdll/unix/signal_arm.c:292: undefined reference to
`_ULarm_get_reg'
collect2: error: ld returned 1 exit status
Makefile:103719: recipe for target 'dlls/ntdll/ntdll.so' failed
make[1]: *** [dlls/ntdll/ntdll.so] Error 1
make[1]: Leaving directory '/home/admin/wine-6.22'
Makefile:13: recipe for target 'install' failed
make: *** [install] Error 2

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #18 from David Paradis  ---
Because the .configure is not finding and dealing with unwind it is setting the
UNWIND_LIBS library define to empty.  Even though I have passed in the location
of the lib on configure it still seems to be ignoring it.

I am not strong on the whole command line make and how all of the checks get
done to determine if the unwind library is present.  It appears that the
.configure is attempting to compile and include unwind and fails, therefore
disabling it.  As far as I know I have installed all of the pieces for
libunwind-dev and verified that they are present in the /usr/include and lib
folders.

At this point I am kind of stumped as to what else to try.

Again, thank you for the assistance

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #19 from Martin Storsjö  ---
Can you clarify what distribution (+version) and architecture configuration of
it that you're building on?

The core issue shown in the configure log is this:

configure:16584: checking for unw_step
configure:16602: gcc -o conftest -g -O2 -mthumb -mfloat-abi=hard    conftest.c
> &5

/tmp/cc8aT0B2.o: In function `main':
/home/admin/wine-6.22/conftest.c:112: undefined reference to `_ULarm_step'
collect2: error: ld returned 1 exit status

On an aarch64 install of Ubuntu 20.04, I've installed the libunwind-dev:armhf
package, and I configure my build like this:
     CC="arm-linux-gnueabihf-gcc" ../wine/configure --host=arm-linux-gnueabihf

With such a build, with Wine 6.22, my current suite of exception handling tests
pass.

As your build just runs plain "gcc", I would guess that it's a 32 bit
distribution. But that doesn't explain the linking error - it sounds like
libunwind is built/packaged differently on that architecture. (Building
libunwind from source is quite quick and straightforward though.)

However the testcase you've attached here does even when Wine is correctly
configured and built with libunwind, as the __CxxFrameHandler function still
isn't implemented. If you'd make a build that is linked against the static CRT
(/MT instead of /MD) it should work - I tried building it from the source you
provided that way, and then it passes.

I can try to have a look at whether it's feasible to implement that function -
but I'm not sure if there are other architecture specific bits in the
msvcp140.dll and how well that DLL is supported on non-x86 architectures
overall.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #20 from Martin Storsjö  ---
(In reply to Martin Storsjö from comment #19)
> However the testcase you've attached here does even when Wine is correctly
> configured and built with libunwind, as the __CxxFrameHandler function still
> isn't implemented. If you'd make a build that is linked against the static
> CRT (/MT instead of /MD) it should work - I tried building it from the
> source you provided that way, and then it passes.
>
> I can try to have a look at whether it's feasible to implement that function
> - but I'm not sure if there are other architecture specific bits in the
>
> msvcp140.dll and how well that DLL is supported on non-x86 architectures
> overall.

Sorry for the mixup, this function doesn't have anything to do with
msvcp140.dll, it's all msvcrt/ucrtbase.

But in any case, yeah it's unimplemented. Looking at the x86_64 version of it,
it's a pretty significant amount of code - I don't think I can make a
corresponding arm version in the near future. For now, the simplest solution is
to just statically link the CRT for such cases.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #21 from Martin Storsjö  ---
Oh, and for the record, a couple known gotchas:

- Unwinding only works in ELF builds, it doesn't work in PE/mingw builds (as
LLVM can't generate the necessary unwind info for windows)

- If the ELF parts are built with Clang instead of GCC, one has to apply this
fix to libunwind,
https://github.com/libunwind/libunwind/commit/971989687f9cf46d9d7f95df593657...,
or add "-gdwarf-2" to CFLAGS (e.g. as CFLAGS="-g -gdwarf-2 -O2")

- Because ARM primarily uses a different unwind info format (.ARM.extab
sections instead of .eh_frame), libunwind seems to rely on parsing the
.debug_frame sections, thus unwinding doesn't work if the wine .so files are
stripped

- Latest git versions of libunwind since
https://github.com/libunwind/libunwind/commit/a4014f33775321b4106a1134b89020...
seem to fail (haven't yet debugged why)

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #22 from Zebediah Figura  ---
FWIW, the configure script tries first to check for unw_step(), and then tries
it again linking to UNWIND_LIBS (which defaults to -lunwind). You've quoted the
first check, but both fail in the log, and I'm guessing it's the latter that's
supposed to work.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #23 from Martin Storsjö  ---
(In reply to Zebediah Figura from comment #22)
> FWIW, the configure script tries first to check for unw_step(), and then
> tries it again linking to UNWIND_LIBS (which defaults to -lunwind). You've
> quoted the first check, but both fail in the log, and I'm guessing it's the
> latter that's supposed to work.

Oh, indeed. For some reason, the second test doesn’t print any executed command
in this log:

configure:16618: checking for unw_step in libunwind
configure:16645: result: no

In my logs, I see the successful command here.

Does --with-unwind affect this somehow? (I don’t pass that explicitly, I just
let it autodetect it.)

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #24 from David Paradis  ---
(In reply to Martin Storsjö from comment #23)
> (In reply to Zebediah Figura from comment #22)
> > FWIW, the configure script tries first to check for unw_step(), and then
> > tries it again linking to UNWIND_LIBS (which defaults to -lunwind). You've
> > quoted the first check, but both fail in the log, and I'm guessing it's the
> > latter that's supposed to work.
>
> Oh, indeed. For some reason, the second test doesn’t print any executed
> command in this log:
>
> configure:16618: checking for unw_step in libunwind
> configure:16645: result: no
>
> In my logs, I see the successful command here.
>
> Does --with-unwind affect this somehow? (I don’t pass that explicitly, I
> just let it autodetect it.)

Thank you to everyone that has assisted with this issue.  I will try to
summarize what I have done.  BTW, with all of the suggestions and my kind of
hacking away I have been able to get my test application to successfully catch
and handle the exception.

This is a remote device that is installed at retail stores across the US and is
running a custom Debian linux distro built thru Yacto (I did not build the OS)
running on a beagle bones ARM32 architecture.

What I had to do was this:
1) Run the apt install libunwind-dev to get the libs and includes as previously
mentioned.
2) Run ./configure with the original set of arguments that I posted on previous
comment.  In looking thru the configure process, the "unwind" is enabled by
default so that it can see if it is available on the system.
3) After the configure, I edited the /include/config.h.in file to comment out
the "#undef HAVE_LIBUNWIND" and replace it with "#define HAVE_LIBUNWIND"
4) Then edit the /dlls/ntdll/Makefile.in to add to the EXTRALIBS line the path
to the unwind lib at "/usr/lib/arm-linux-gnueabihf/libunwind.so
"
5) Then run make and install from the /dlls/ntdll folder
6) Finally, in my project I switched from /MD to /MT for the CRT linking.

Now when I run my program everything works as expected.  If I do not do step 3
and 4 then I get the prior messages about libunwind not being available when
the exception is thrown in my application.

I don't know why the UNWIND_LIBS variable is not set correctly to properly
include the libunwind.so, but if the ./configure process correctly detected my
installed libunwind and set the .so path location correctly then I wouldn't
have to modify any of the wine files at all.  For now, I am ok with doing my
own internal documentation for future builds to be make sure that I have these
steps documented in the event I have to do this again in the future.

Again, thank you all for the help, it is greatly appreciated.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #25 from David Paradis  ---
I just wanted to give a quick update now that I have moved forward to
production code.

I started to modify my real production code to switch the static CRT library,
but unfortunately that is causing my some issues because I have a 3rd party
library that using /MD switches and when I attempt to compile my code against
that library I have issues unresolved external symbols that are being used in
that library related to things like fopen, srand and other MSVCRT functions.

I am going to keep trying to change compiler switches on my project and the 3rd
party library, but as of now the requirement to use static linked CRT might be
a no go for my solution.

[Bug 50625] Unhandled exception with try/catch

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=50625

--- Comment #26 from David Paradis  ---
Let me first again thank you all for the assistance with this issue, it is
greatly appreciated.  Makes me extremely happy that this community is active
and addressing issues.

I have successfully been able to upgrade and build the Wine 6.22 sources and
all of my companies packages with the CRT as static rather than dynamic. I have
also been able to force the 3rd party code to build statically as well and all
is now working correctly on my platform.  Exceptions thrown in the 3rd party
library is being caught successfully in my code.

So, as far as I am concerned this issue is addressed when the projects are
configured properly and you have the libunwind-dev package installed.

I had to do the following steps before building Wine on my box in order for it
to detect and build against the Libunwind package:
  * After running the "./configure", edit the "/include/config.h.in" to replace
the "#undef HAVE_LIBUNWIND" with "#define HAVE_LIBUNWIND".
  * Edit the "dlls/ntdll/Makefile.in" to add to the end of the "EXTRALIBS"
definition the path libunwind of "/usr/lib/arm-linux-gnueabihf/libunwind.so"
  * Then on my project side, ensure that the CRuntimeLibrary is static rather
than dynamic.