Support for precise stepping in functions compiled before debugging was started (step 1)

Support for precise stepping in functions compiled before debugging was started (step 1)

This change will ensure that all non-optimized code will be compiled
with debug break slots when debugging is initiated. This is handled by
scanning the heap for non-optimized functions without debug break slots and setting their code to be lazy recomplied. When the lazy recompilation happens the code will ge generated with debug break slots (if debugging is still active at that point in time).

R=svenpanne@chromium.org
Currently this is only implemented for functions which do not have activations on the stack.

BUG=
TEST=

Committed: http://code.google.com/p/v8/source/detail?r=9250

[Sven Panne, 2011-09-08 08:07:59]

http://codereview.chromium.org/7839030/diff/2001/src/debug.cc
File src/debug.cc (right):

http://codereview.chromium.org/7839030/diff/2001/src/debug.cc#newcode1686
src/debug.cc:1686: bool SortedListContains(List<JSFunction*>* functions,
JSFunction* function) {
Perhaps this should be moved to the List template class, probably with a
different signature (something like C's bsearch)

http://codereview.chromium.org/7839030/diff/2001/src/debug.cc#newcode1720
src/debug.cc:1720: List<JSFunction*> functions(100);
functions => active_functions

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h
File src/objects-inl.h (right):

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h#newcode2950
src/objects-inl.h:2950: kFullCodeFlagsHasDeoptimizationSupportShift) == 1;
Why do we need both a mask and a shift? A single bit mask/flag (whatever you
call it) seems to be sufficient. Same for other places like this below.

[Søren Thygesen Gjesse, 2011-09-13 07:31:18]

PTAL

http://codereview.chromium.org/7839030/diff/2001/src/debug.cc
File src/debug.cc (right):

http://codereview.chromium.org/7839030/diff/2001/src/debug.cc#newcode1686
src/debug.cc:1686: bool SortedListContains(List<JSFunction*>* functions,
JSFunction* function) {
On 2011/09/08 08:07:59, Sven wrote:
> Perhaps this should be moved to the List template class, probably with a
> different signature (something like C's bsearch)

Added the following template functions to list.h. I decided against adding
bsearch directly to list as I did not want to add a flag for bookkeeping whether
the list is currently sorted.

// Perform binary search for an element in an already sorted
// list. Returns the index of the element of -1 if it was not found.
template <typename T>
int SortedListBSearch(List<T>& list, T elem, int (*cmp)(const T* x, const T*
y));
template <typename T>
int SortedListBSearch(List<T>& list, T elem);

http://codereview.chromium.org/7839030/diff/2001/src/debug.cc#newcode1720
src/debug.cc:1720: List<JSFunction*> functions(100);
On 2011/09/08 08:07:59, Sven wrote:
> functions => active_functions

Done.

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h
File src/objects-inl.h (right):

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h#newcode2950
src/objects-inl.h:2950: kFullCodeFlagsHasDeoptimizationSupportShift) == 1;
On 2011/09/08 08:07:59, Sven wrote:
> Why do we need both a mask and a shift? A single bit mask/flag (whatever you
> call it) seems to be sufficient. Same for other places like this below.

This pattern is used for all the other flags (at lease for class Code), as some
flags are more than one bit. However I changed the initialization of the Mask
values in objects.h to

  static const int kFullCodeFlagsHasDeoptimizationSupportMask =
      1 << kFullCodeFlagsHasDeoptimizationSupportShift;
  static const int kFullCodeFlagsHasDebugBreakSlotsMask =
      1 << kFullCodeFlagsHasDebugBreakSlotsShift;

[Sven Panne, 2011-09-13 07:51:51]

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h
File src/objects-inl.h (right):

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h#newcode2950
src/objects-inl.h:2950: kFullCodeFlagsHasDeoptimizationSupportShift) == 1;
On 2011/09/13 07:31:19, Søren Gjesse wrote:
> On 2011/09/08 08:07:59, Sven wrote:
> > Why do we need both a mask and a shift? A single bit mask/flag (whatever you
> > call it) seems to be sufficient. Same for other places like this below.
> 
> This pattern is used for all the other flags (at lease for class Code), as
some
> flags are more than one bit. However I changed the initialization of the Mask
> values in objects.h to
> 
>   static const int kFullCodeFlagsHasDeoptimizationSupportMask =
>       1 << kFullCodeFlagsHasDeoptimizationSupportShift;
>   static const int kFullCodeFlagsHasDebugBreakSlotsMask =
>       1 << kFullCodeFlagsHasDebugBreakSlotsShift;

I think Kevin changed Code to use BitField for flags
(http://codereview.chromium.org/7787028/), so we should probably follow that
style.

[Søren Thygesen Gjesse, 2011-09-13 08:06:29]

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h
File src/objects-inl.h (right):

http://codereview.chromium.org/7839030/diff/2001/src/objects-inl.h#newcode2950
src/objects-inl.h:2950: kFullCodeFlagsHasDeoptimizationSupportShift) == 1;
On 2011/09/13 07:51:51, Sven wrote:
> On 2011/09/13 07:31:19, Søren Gjesse wrote:
> > On 2011/09/08 08:07:59, Sven wrote:
> > > Why do we need both a mask and a shift? A single bit mask/flag (whatever
you
> > > call it) seems to be sufficient. Same for other places like this below.
> > 
> > This pattern is used for all the other flags (at lease for class Code), as
> some
> > flags are more than one bit. However I changed the initialization of the
Mask
> > values in objects.h to
> > 
> >   static const int kFullCodeFlagsHasDeoptimizationSupportMask =
> >       1 << kFullCodeFlagsHasDeoptimizationSupportShift;
> >   static const int kFullCodeFlagsHasDebugBreakSlotsMask =
> >       1 << kFullCodeFlagsHasDebugBreakSlotsShift;
> 
> I think Kevin changed Code to use BitField for flags
> (http://codereview.chromium.org/7787028/), so we should probably follow that
> style.

I also noticed that after the rebase. This has now been changed.

[Sven Panne, 2011-09-13 08:11:41]

lgtm