[lldb-dev] lldb.frame.EvaluateExpression slows down when called a lot

[Scott Knight]

Thanks for the information Greg. I have a quick followup. I'm using the
version of lldb that comes with XCode 5.1.1

Launching it like this

Scotts-MacBook-Pro:~ scottknight$
/Applications/Xcode.app/Contents/Developer/usr/bin/lldb -p 13892
Attaching to process with:
    process attach -p 13892
Process 13892 stopped
Executable module set to
"/Users/scottknight/.rbenv/versions/2.1.1/bin/ruby".
Architecture set to: x86_64-apple-macosx.

When I tried using GetValueForVariablePath I got 'No value' back. See the
output below.

(lldb) v
lldb-310.2.37
(lldb) script
Python Interactive Interpreter. To exit, type 'quit()', 'exit()' or Ctrl-D.
>>> print lldb.frame.EvaluateExpression('ruby_current_vm')
(rb_vm_t *) $1 = 0x00007f9a01003000
>>> print lldb.frame.GetValueForVariablePath('ruby_current_vm')
No value

Since ruby_current_vm is a global variable is there something different I
would need to do to access it?

Thanks,
Scott Knight


On Thu, Apr 17, 2014 at 1:45 PM, Greg Clayton <gclayton at apple.com> wrote:

> Running expressions has all sorts of side effects like storing data in the
> inferior program and it also involves running the clang expression parser
> which can be expensive.
>
> You can, from a frame, get a SBValue for a variable without using the
> expression parser:
>
>     lldb::SBValue
>     SBFrame.GetValueForVariablePath (const char *var_path);
>
> So you can change your code to this:
>
> heaps_used =
> lldb.frame.GetValueForVariablePath('ruby_current_vm->objspace->heap_pages.used').GetValueAsUnsigned(0)
>
> for i in xrange(heaps_used):
>     page =
> lldb.frame.GetValueForVariablePath('*ruby_current_vm->objspace->heap_pages.sorted[%i]'
> % i)
>
> The GetValueForVariablePath() will find the variable and not create a
> temporary each time. It also doesn't use the expression parser at all so it
> won't call any code. The objects you access must be available in the
> hierarchy of the struct or class and the struct or class can't override the
> "->" operator. Other than that, the GetValueForVariablePath() knows how to
> access members ("ruby_current_vm->objspace->heap_pages.sorted"),
> dereference pointers using the array syntax ("my_ptr[12]"), deref a pointer
> ("*this->other_ptr"), and take the address of something
> ("&ruby_current_vm->objspace->heap_pages.sorted[12]").
>
> So give the GetValueForVariablePath a try. The SBValue returned is
> something that represents the live variable value, not a const result like
> you get back from expression. SBValue you get back is tied to the frame
> from which you got it, so it will continue to evaluate correctly and its
> value will change if you step between calling functions with it. If the
> frame it came from goes away (step out), then it won't return any valid
> values again as it will detect the frame is gone and stop answering any
> questions. So you should always fetch a fresh value from the frame each
> time you want to use it.
>
> Greg
>
>
> On Apr 17, 2014, at 7:57 AM, Scott Knight <knightsc at gmail.com> wrote:
>
> > I was recently using lldb to connect to a debug build of ruby to inspect
> the heap. In order to do this I was doing something like this
> >
> > -----------
> > heaps_used =
> lldb.frame.EvaluateExpression('ruby_current_vm->objspace->heap_pages.used').GetValueAsUnsigned(0)
> >
> > for i in xrange(heaps_used):
> >     page =
> lldb.frame.EvaluateExpression('*ruby_current_vm->objspace->heap_pages.sorted[%i]'
> % i)
> > -----------
> >
> > What I noticed was that for each EvaluateExpression a temporary $0, $1,
> $2, etc.. variable is created. If I ended up calling my python code
> multiple times more and more variables seemed to pile up and every
> EvaluateExpression call seemed to take longer and longer.
> >
> > I tried calling EvaluateExpression how I would call expr from the lldb
> command line setting my own variable, so something like
> >
> > lldb.frame.EvaluateExpression('int $test = 5')
> >
> > But that seemed to error out. So is there some other way in the API that
> is better for accessing global variables that won't slow down. Is this
> something actually wrong with the debugger? I can create an actual test
> case similar to the test suite in lldb if that would be helpful.
> >
> > Thanks,
> > Scott Knight
> >
> > _______________________________________________
> > lldb-dev mailing list
> > lldb-dev at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20140417/760b055d/attachment.html>