Re: [Jikesrvm-researchers] Using jdp on IA32

[Janice S. <ja...@us...>]

Hi,


1) While I'm gathering together a list of items within jdp that need work,
that list is not complete yet. The commands I know of that are not yet
supported on Lintel are:  listt with the run or system options; stepbr;
stepline (as you observed);  steplineover; and cthread;
As well, we are aware that some commands that are implemented sometimes
don't work as expected (that applies to both Lintel and AIX).  When I
discover cases I'm adding them to my work list.
Also, we can see a need to expand on the friendliness of jdp by adding more
functionality such as giving a list of method choices when an overloaded
methodname is found and supporting some simple arithmetic with some
commands.

Have you tried the whereframe command?   Perhaps that will help you with
your changing stack frame concern. (Sorry, I haven't had a chance to test
it out first).  Or do you mean change stackframes between threads?

I'll correct the documentation about LoadLocalVariableTables. I'll look
into why this is not sufficient to get the debugger to know about the
variables.  Yes, shadow is a term we started using years ago. It just means
a local copy where you may or may not have made some private modifications.


2) Classes that are part of the bootImage have already had their class
initializers run as part of building the bootImage. So those classes are
marked already initialized.  At times we need more initialization to happen
at the start of running the bootImage, then we put the extra initialization
in a "boot( )" method (historical naming convention), and arrange for that
boot method to be called from VM.boot( ).


3)  If you are getting a segmentation fault followed by a "jid>" prompt,
then it sounds more like you are crashing the debugger. If the segmentation
fault is followed by  a jdp prompt then perhaps this is showing a handled
case, but in my experience the exceptions that are more likely to be
handled don't indicate segmentation fault.  I'm not sure what you mean by
distinguishing between handled and unhandled execptions.  Do you mean is
there a jdp command to ask if the exception has a handler in the user app
or JikesRVM?  No, I don't think there is such a command.

4) The where and whereframe commands take an optional argument of <from |
to> <hexval>
   While I have not had a chance to try this, it would appear to allow you
to look beyond the last 20 frames.  If you knew the start of the thread's
stack then you could request the where to start from there.  I'll add a
command to learn about such thread details to the list of possible new
functions to be added to jdp.

5)  Dave Grove indicated he'll be looking at this

Thanks for the input on jdp.

Janice
-----------------------------------------------------------------------------

Janice C. Shepherd
 ja...@wa... (or ja...@us...)
IBM T.J. Watson Research
1S-B57 Hawthorne
Phone: (914)784-7571  tieline: 863-7571



John Leuner <je...@pi...>@www-124.southbury.usf.ibm.com on 02/16/2002
05:33:43 PM

Please respond to jik...@ww...

Sent by:    jik...@ww...


To:    jik...@ww...
cc:
Subject:    [Jikesrvm-researchers] Using jdp on IA32



I started playing with jdp for the first time yesterday on IA32 Linux.

1)

I noticed that the sl command is unimplemented, is there a list somewhere
of TODO work on JDP?

Also, I don't see a way to change to a different stack frame and examine
data there.

Whenever I use the p command, it always tells me there are no local
variables available.

The Debugging Tips section of the User's Guide says this:

    * Compiling the system with local variables: The default boot image is
    built without the local variable tables. To get local variables, set
    this flag to true in your local shadow:

VM_Control.LoadLocalVariableTables

      Then rebuild the boot image. This will increase the boot image size
      by about 10%.


I assume that this "shadow" terminology refers to prior JRVM build systems?

There is no VM_Control.java in the RVM.classes directory, but VM_Properties
does have a LoadLocalVariableTables field.

I have set this to true, but I can't see any difference.

I am compiling with BaseBaseSemispace.

2) I was making changes to the VM_CommandLineArgs.java file. I noticed that
the static initializer (which should print out the prefix list when DEBUG
is true) only runs during the build process. When JikesRVM is invoked the
static initializer isn't run. Is this related to the 'tool' mode that I've
heard people talking about? Is it legal to run methods in a class without
invoking the static initializer first?

3) I have noticed that sometimes I can run something successfully by just
invoking JikesRVM, but when I run the same thing in jdp, it will stop with
a segmentation fault. I'm guessing that this will be handled by a handler
and the program will continue running. Is there a way to distinguish
between 'handled' and 'unhandled' signals?

4) I have some situations where the system is recursing and then faulting.
jdp only shows me the last 20 frames, but I am actually interested in the
top 20 frames, how can I see these?

5) When Jikes encounters a native method call that it can't satisfy, I see
the message VM_Runtime.unexpectedNativeMethodCall. How do I determine which
method was being called in which class?

John Leuner

_______________________________________________
Jikesrvm-researchers mailing list
Jik...@ww...
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-researchers