Smalltalk › Cincom › VisualWorks

[BUG] crash when generating code

‹ Previous Topic Next Topic ›
 
classic Classic list List threaded Threaded
♦
♦
 4 messages Options
Christian Haider-2
Reply | Threaded
Open this post in threaded view
♦
♦
|

[BUG] crash when generating code

Christian Haider-2
Hi,
 
recently the vw is crashing sometimes, but not always, with the following message (from the debug vm):
------------------------------
Assert fail @ 312 in stack\send.c       isTranFailCode(nm)
VisualWorks(R) 7.4d Oct 24 2006
[ thapi com ephemerons immutability debug assert ]
Fatal error: unexpected translation failure (selector 'ihiDAX'), code = 2
 
Space could not be allocated.
------------------------------
 
Context:
WinXP sp2, Intel Core2Duo 6300, 2 GB, VW 7.4.1
 
The code retrieves data from Bloomberg and with the result I compile a method (in this case #ihiDAX). This results in the crash. Unfortunately this is not deterministic, since this succeeds most of the time. Also, in this case the generated method is not big (sometimes I get methods with a couple of megebytes source without problems).
 
Also, with this particular case, the image was just startet, but I get it also after many hours of work after lots of requests which succeed. And, yes, I do have enough memory in my box.
 
cheers,
    Christian
 
 
Smalltalked Visuals GmbH
Christian Haider
Wiesbaden, Germany
 
Reinout Heeck-2
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: [BUG] crash when generating code

Reinout Heeck-2
Christian, did you see Joachim Geidel's message on this subject (on the
18th)? He reports


   ... which
   attempts to flush the method cache for the new method's
   selector by using #flushVMmethodCacheEntriesFor:.
   Apparently, this is not sufficient.



That suggests you need to experiment a bit with how you flush the code
cache - see if it works better if you use the more expensive
#flushVMmethodCache instead.


R
-



> Hi,
>  
> recently the vw is crashing sometimes, but not always, with the
> following message (from the debug vm):
> ------------------------------
> Assert fail @ 312 in stack\send.c       isTranFailCode(nm)
> VisualWorks(R) 7.4d Oct 24 2006
> [ thapi com ephemerons immutability debug assert ]
> Fatal error: unexpected translation failure (selector 'ihiDAX'), code = 2
>  
> Space could not be allocated.
> ------------------------------
>  
> Context:
> WinXP sp2, Intel Core2Duo 6300, 2 GB, VW 7.4.1
>  
> The code retrieves data from Bloomberg and with the result I compile a
> method (in this case #ihiDAX). This results in the crash. Unfortunately
> this is not deterministic, since this succeeds most of the time. Also,
> in this case the generated method is not big (sometimes I get methods
> with a couple of megebytes source without problems).
>  
> Also, with this particular case, the image was just startet, but I get
> it also after many hours of work after lots of requests which succeed.
> And, yes, I do have enough memory in my box.
>  
> cheers,
>     Christian
>  
>  
> Smalltalked Visuals GmbH
> Christian Haider
> Wiesbaden, Germany
> http://www.smallCharts.de <http://www.smallcharts.de/>
>  

Christian Haider-2
Reply | Threaded
Open this post in threaded view
♦
♦
|

AW: [BUG] crash when generating code

Christian Haider-2
In reply to this post by Christian Haider-2
Sorry for the false alarm - forunately it is not the code generation part!
 
The bug (if it would qualify as one) is with executing a method with a couple of cascades (a lot :-).
 
Attached is a method with reproduces the problem. If anyone is interested, file-in the method in a virgin visual.im.
Executing it (with a doit on the method comment) will succeed (on my machine).
If you scroll down to "(Date newDay: 13 monthNumber: 5 year: 2005) asTimestamp" you'll find a comment with some explanantion. The rest of the cascade is commented out. If you uncomment one more #add: statement, executing the method will crash the image (on my machine - the crash provoking number of statements might be different on your machine).
 
Interestingly, if you you just select the whole thing (except for the method selector), and inspect-it, you will get a nice error notification with a primitive failed.
 
Obviously, I hit a vm limit for cascades (for this method on my machine it is 2081 - 2082 will exceed the stack space and crash). The limit depends on the concrete code; the original method had a few more method calls and crashed at 2054.
 
This behavior is not too critical for me, but just good to know.
 
The bug, in my opinion, is that the vm crashes. It would be so much nicer if it could also raise a primitive failed like it does when interpreting the code.
 
Cheers,
    Christian

Von: Christian Haider
Gesendet: Freitag, 23. Februar 2007 10:16
An: [hidden email]
Betreff: [BUG] crash when generating code

Hi,
 
recently the vw is crashing sometimes, but not always, with the following message (from the debug vm):
------------------------------
Assert fail @ 312 in stack\send.c       isTranFailCode(nm)
VisualWorks(R) 7.4d Oct 24 2006
[ thapi com ephemerons immutability debug assert ]
Fatal error: unexpected translation failure (selector 'ihiDAX'), code = 2
 
Space could not be allocated.
------------------------------
 
Context:
WinXP sp2, Intel Core2Duo 6300, 2 GB, VW 7.4.1
 
The code retrieves data from Bloomberg and with the result I compile a method (in this case #ihiDAX). This results in the crash. Unfortunately this is not deterministic, since this succeeds most of the time. Also, in this case the generated method is not big (sometimes I get methods with a couple of megebytes source without problems).
 
Also, with this particular case, the image was just startet, but I get it also after many hours of work after lots of requests which succeed. And, yes, I do have enough memory in my box.
 
cheers,
    Christian
 
 
Smalltalked Visuals GmbH
Christian Haider
Wiesbaden, Germany
 

Object class-ihiDAXPrimitiveFailed.st (393K) Download Attachment
Eliot Miranda-2
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: [BUG] crash when generating code

Eliot Miranda-2


On 2/23/07, Christian Haider <[hidden email]> wrote:
Sorry for the false alarm - forunately it is not the code generation part!
 
The bug (if it would qualify as one) is with executing a method with a couple of cascades (a lot :-).
 
Attached is a method with reproduces the problem. If anyone is interested, file-in the method in a virgin <a href="http://visual.im" target="_blank" onclick="return top.js.OpenExtLink(window,event,this)">visual.im.
Executing it (with a doit on the method comment) will succeed (on my machine).
If you scroll down to "(Date newDay: 13 monthNumber: 5 year: 2005) asTimestamp" you'll find a comment with some explanantion. The rest of the cascade is commented out. If you uncomment one more #add: statement, executing the method will crash the image (on my machine - the crash provoking number of statements might be different on your machine).
 
Interestingly, if you you just select the whole thing (except for the method selector), and inspect-it, you will get a nice error notification with a primitive failed.
 
Obviously, I hit a vm limit for cascades (for this method on my machine it is 2081 - 2082 will exceed the stack space and crash). The limit depends on the concrete code; the original method had a few more method calls and crashed at 2054.
 
This behavior is not too critical for me, but just good to know.
 
The bug, in my opinion, is that the vm crashes. It would be so much nicer if it could also raise a primitive failed like it does when interpreting the code.


There is an AR for this.  What's happening is that the generated native method is too large to fit in the native code cache.  Try increasing the size (see ObjectMemory class>>setSizesAtStratup:).  Before I left Steve Dahl and I had discussed that a good response would be for the translator to always fail to translate methods above a certain size (say 5k bytecodes) and cause a message to be sent to the context, e.g. Context>>methodTooLargeToTranslate, and have contxts respiond to this by interpreting their method.


Cheers,
    Christian

Von: Christian Haider
Gesendet: Freitag, 23. Februar 2007 10:16
An: [hidden email]
Betreff: [BUG] crash when generating code

Hi,
 
recently the vw is crashing sometimes, but not always, with the following message (from the debug vm):
------------------------------
Assert fail @ 312 in stack\send.c       isTranFailCode(nm)
VisualWorks(R) 7.4d Oct 24 2006
[ thapi com ephemerons immutability debug assert ]
Fatal error: unexpected translation failure (selector 'ihiDAX'), code = 2
 
Space could not be allocated.
------------------------------
 
Context:
WinXP sp2, Intel Core2Duo 6300, 2 GB, VW 7.4.1
 
The code retrieves data from Bloomberg and with the result I compile a method (in this case #ihiDAX). This results in the crash. Unfortunately this is not deterministic, since this succeeds most of the time. Also, in this case the generated method is not big (sometimes I get methods with a couple of megebytes source without problems).
 
Also, with this particular case, the image was just startet, but I get it also after many hours of work after lots of requests which succeed. And, yes, I do have enough memory in my box.
 
cheers,
    Christian
 
 
Smalltalked Visuals GmbH
Christian Haider
Wiesbaden, Germany
<a href="http://www.smallcharts.de/" target="_blank" onclick="return top.js.OpenExtLink(window,event,this)">http://www.smallCharts.de
 
Free forum by Nabble Edit this page