Hi,
Sorry to have to go on with this... but I do want
my built from source FG to work!
Anyway, I 'think' I have found a JSB 'bug', or at
least something I do not understand...
In FGStandardAtmosphere bind(void) we have 2 typedef
typedef double (FGStandardAtmosphere::*PMFi)(int) const;
typedef void (FGStandardAtmosphere::*PMF)(int, double);
And then we have 2 functions GetTemperatureBias, and
Set TemperatureBias, which are declared similarly to the
typedefs, and tied -
PropertyManager->Tie("atmosphere/delta-T", this, eRankine,
(PMFi)&FGStandardAtmosphere::GetTemperatureBias,
(PMF)&FGStandardAtmosphere::SetTemperatureBias);
That is the getter declaration matches the function
declaration...
But the 3rd tie in the bind() is -
PropertyManager->Tie("atmosphere/P-sl-psf", this, ePSF,
(PMFi)&FGStandardAtmosphere::GetPressureSL,
(PMF)&FGStandardAtmosphere::SetPressureSL);
How come GetPressureSL is defined as -
virtual double GetPressureSL(void) const
{ return SLpressure }
Here we have a function declared differently to what
the typedef of the indexed getter pointer is...
It is the call to this 'getter' that crashes my
system!
The rest of this is just how I tried to track this down,
and you can skip it if not in Windows, or just not
interested ;=)) just the big tedium of bug chasing...
To rule out the Win 7 system in general downloaded and
installed FG 2.8.0-RC4, and it worked perfectly. This time
fgrun had all the correct paths ;=)) Thanks Fred...
Maybe I should head towards building the 64-bit
exe, rather than this problematic 32-bit first ;=()
I then increased log level to debug, and it seems to
consistently break after the output shown, but as we
know this does not always really indicate where the
real problem is...
...
Tied /ai[0]/models[0]/carrier[1]/controls[0]/turn-to-recovery-hdg[0]
Tied /ai[0]/models[0]/carrier[1]/controls[0]/turn-to-base-course[0]
Submodels: Loading AI submodels
AI Manager: AI model return list size 2
Submodels: Loading sub submodels
trigger listener #88
trigger listener #93
See full log:
http://geoffair.org/tmp/templog.txt
I tried adding --disable-ai-traffic --disable-ai-scenarios
even --disable-ai-models, but that had no effect ;=((
See full log:
http://geoffair.org/tmp/templog5.txt
So I built the MSVC10 Debug version, and ran fgfs under
the MSVC debugger, and it certainly 'aborts' at line 589
in props.hxx, but it is during the saveInitialState()
call in fg_init.cxx... in copyProperties(...)... so
seems nothing to do with AI...
At the moment of the crash, the call stack is -
> fgfs.exe!
SGRawValueMethodsIndexed<JSBSim::FGStandardAtmosphere,double>::getValue() Line
589 + 0x27 bytes C++
fgfs.exe!SGPropertyNode::get_double() Line 396 + 0x1b bytes C++
fgfs.exe!SGPropertyNode::getDoubleValue() Line 1192 + 0x8 bytes
C++
fgfs.exe!copyProperties(const SGPropertyNode * in, SGPropertyNode *
out, int attr_value, int attr_mask) Line 672 + 0x8 bytes C++
fgfs.exe!copyProperties(const SGPropertyNode * in, SGPropertyNode *
out, int attr_value, int attr_mask) Line 732 + 0x24 bytes C++
fgfs.exe!copyProperties(const SGPropertyNode * in, SGPropertyNode *
out, int attr_value, int attr_mask) Line 732 + 0x24 bytes C++
fgfs.exe!copyProperties(const SGPropertyNode * in, SGPropertyNode *
out, int attr_value, int attr_mask) Line 732 + 0x24 bytes C++
fgfs.exe!copyProperties(const SGPropertyNode * in, SGPropertyNode *
out, int attr_value, int attr_mask) Line 732 + 0x24 bytes C++
fgfs.exe!FGGlobals::saveInitialState() Line 414 + 0x25 bytes C++
fgfs.exe!fgInitSubsystems() Line 1316 C++
fgfs.exe!fgIdleFunction() Line 400 + 0x5 bytes C++
fgfs.exe!fgOSMainLoop() Line 284 + 0x5 bytes C++
fgfs.exe!fgMainInit(int argc, char * * argv) Line 537 + 0x5 bytes
C++
fgfs.exe!main(int argc, char * * argv) Line 251 + 0xd bytes C++
fgfs.exe!__tmainCRTStartup() Line 555 + 0x19 bytes C
fgfs.exe!mainCRTStartup() Line 371 C
kernel32.dll!@BaseThreadInitThunk@12() + 0x12 bytes
ntdll.dll!___RtlUserThreadStart@8() + 0x27 bytes
ntdll.dll!__RtlUserThreadStart@8() + 0x1b bytes
Tediously tracing through some code note the following...
In props_io.cxx, while iterating nChildren = 2,
Ln 705 for(int i = 0; i < nChildren; i++), on the
first child, i == 0, the in_child _name="jsbsim",
out_child="jsbsim", attr_value=3, attr_mask=195
The multiple copyProperties() is just as it
iterates through children of children of children...
Ln 672: case props::DOUBLE:
if (!out->setDoubleValue(in->getDoubleValue()))
retval = false;
break;
Both in and out = "fdm"
In props.cxx getDoubleValue()
// Shortcut for common case
if (_attr == (READ|WRITE) && _type == props::DOUBLE)
Ln 1192: return get_double()
Again in props.cxx get_double ()
if (_tied)
Ln 396: return static_cast<SGRawValue<double>*>
(_value.val)->getValue();
which put me in the 'problem' _getter
fgfs.exe!SGRawValueMethodsIndexed<JSBSim::
FGStandardAtmosphere,double>::getValue() Line 589 + 0x27 bytes
getting the tied <JSBSim::FGStandardAtmosphere,double>
By setting a debug trap here the first call to here is to
get_magnetos(int engine) in src\aircraft\controls.hxx,
which I can see in disassembly is a stdcall. That is
it ends with ret 4!
The next call to this 589 is get_carb_heat, again a
stdcall - ends with ret 4!
But there are __cdecl on some things, like
call SGAnimation::animate
add esp,18h ; fix up the stack
So, the code is a mixture of __cdecl and stdcall!
But who, what, how, is this decided? Not sure...
Now to try to back up to where JSBSim is setting
this _getter...
This seems to be in -
src\FDM\JSBSim\models\atmosphere\FGStandardAtmosphere.cpp
v 1.21 2012/04/13 13:18:27 jberndt, so does not appear
to be exactly 'new' code...
In bind(), the getter/setter are tied to functions
void FGStandardAtmosphere::bind(void)
{
typedef double (FGStandardAtmosphere::*PMFi)(int) const;
typedef void (FGStandardAtmosphere::*PMF)(int, double);
PropertyManager->Tie("atmosphere/delta-T", this, eRankine,
(PMFi)&FGStandardAtmosphere::GetTemperatureBias,
(PMF)&FGStandardAtmosphere::SetTemperatureBias);
PropertyManager->Tie("atmosphere/SL-graded-delta-T", this, eRankine,
(PMFi)&FGStandardAtmosphere::GetTemperatureDeltaGradient,
(PMF)&FGStandardAtmosphere::SetSLTemperatureGradedDelta);
PropertyManager->Tie("atmosphere/P-sl-psf", this, ePSF,
(PMFi)&FGStandardAtmosphere::GetPressureSL,
(PMF)&FGStandardAtmosphere::SetPressureSL);
}
These are all 'virtual' functions in FGStandardAtmosphere.h,
replacing the base FGAtmosphere class, FGAtmosphere.h,
which in turn replace the base.base FGModel class, FGModel.h,
in the base.base.base FGModelFunctions, in the basex4
FGJSBBase class... but it all looks fine ;=()
But again and again I see some things are definitely
__cdecl, like -
push eax
push ecx
push edx
push eax
call copyProperties
add esp,10h
And others are stdcall, like
push eax
call get_throttle ; inline get_throttle(int eng)
ret 4
Like all the FGControls - controls.hxx - all end
with ret 4 - stdcall...
BUT, BUT, BUT, WAIT, WAIT, WAIT the function -
virtual double GetPressureSL(void)
is NOT... it does a simple return...
All the others seem be an 'index' getter, but
note GetPressureSL is a (void) - NO INDEX!!!
So WHY does this GetPressureSL get included in
the above bind() where the typedef is -
typedef double (FGStandardAtmosphere::*PMFi)(int)
const;
Is this the problem? It seems to match the dialog
report that a function pointer and a function have
different declarations...
BAH! This seems like an endless chase ;=(( I suppose
I just know, understand enough to get myself into
trouble ;=))
Unless I have somehow stumbled on a real BUG, and
it can be fixed, or someone else has some ideas I
will give it up for a while... this has already
consumed most of today...
Meantime will continue with the big setup to be able
to eventually build the 64-bit version... first
re-compiling all the dependencies in 64-bits...
yuck...
Regards,
Geoff.
-------- Forwarded Message --------
> From: Geoff McLane <[email protected]>
> To: flightgear-devel <[email protected]>
> Subject: Windows Abort Retry Ignore dialog at startup
> Date: Wed, 01 Aug 2012 20:38:42 +0200
>
> Hi all,
>
> Each, SG/FG/fgdata updated to release/2.8.0 today.
> Full SG then FG clean and build using MSVC10 in Win 7,
> 64-bit...
>
> That is with like -
> $ git pull origin next
> $ git fetch --all
> $ git checkout release/2.8.0
> $ git reset --hard origin/release/2.8.0
> $ git branch
> next
> * release/2.8.0
> followed by a full clean and build in an
> o-o-s folder...
>
> When the splash screen is showing 'initializing
> subsystems' after several seconds, get an ugly
> Abort/Retry/Ignore dialog ;=((
>
> See:
> http://geoffair.org/tmp/error-01.jpg
>
> Have tried Ignore, but it keeps coming back,
> and leads to other abort dialogs...
>
> Only console output is -
> KI226 dme indicator #0 initialized
> loading scenerio 'nimitz_demo'
>
> And will also try turning on log level debug
> to see if I can closer track down where and
> when it is happening...
>
> As can be seen in the dialog it indicates the ESP
> was not saved across a function call and that this
> is usually the result of a function declared with
> one calling convention, say stdcall, while a
> function pointer is declared with a different
> convention like say __cdecl ;=((
>
> Is anyone else seeing this? Seen this? Found
> a fix? ...
>
> And indeed the line indicated, 589, in simgear
> prop.hxx is a call through a pointer -
> if (_getter) { return (_obj.*_getter)(_index); }
>
> The getter_t _getter is declared without a
> declared calling convention like -
> typedef T (C::*getter_t)(int) const;
> which means it would be the default, which is
> __cdecl (/Gd) in MSVC10, as it has been for a
> long time...
>
> This suggest somewhere a call setting this _getter
> pointer is passing a function declared with
> say stdcall, fastcall or something other than
> __cdecl...
>
> Anyway, totally at a loss to find and fix this.
>
> Any ideas, thoughts, things to try, very VERY
> welcome...
>
> Regards,
> Geoff.
>
> PS: In not quite the same rigorous cleaning,
> updating re-build etc, also see/seen the SAME
> thing in XP 32-bit with MSVC8... which suggests
> it is not the compiler version, or OS version...
>
> And does NOT happen in Ubuntu 10.04 64-bits...
>
>
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Flightgear-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/flightgear-devel
