The Inbox: System-mt.697.mcz

Is there any scenario, where #home is important for the tally?

Hi,

On 27.01.2015, at 22:21, Levente Uzonyi <[hidden email]> wrote:

> AndreasSystemProfiler

This is not in trunk, right?
If so, why?

AndreasSystemProfiler has the same bug, and it's annoyed me forever, so thanks for the fix.

I'm not an expert, but here's an example:

AndreasSystemProfiler spyOn: [
        [ #((1 2 3)) do: [ :each |
                each findLast: [ :ea |
                        ea squared = ea ] ] ] bench ].

Before the patch you got:

99.98 (5,018)  UndefinedObject  DoIt
  10.39 (521)  Array [SequenceableCollection]  findLast:
   9.68 (486)  Array [SequenceableCollection]  do:
   5.94 (298)  Time class  millisecondClockValue

But the profiled block doesn't send #findLast, nor #do:, and neither #millisecondClockValue. It sends #bench to a block, and nothing else.

After the fix it's:

99.98 (5,018)  UndefinedObject  DoIt
  99.98 (5,018)  BlockClosure  bench
    99.96 (5,017)  UndefinedObject  DoIt
      20.9 (1,049)  Array [SequenceableCollection]  do:
        |16.01 (804)  UndefinedObject  DoIt
        |  16.01 (803)  Array [SequenceableCollection]  findLast:
      5.7 (286)  Time class  millisecondClockValue

Which is much better, and almost exact. #millisecondsClockValue is sent from #bench, not from a block.

Levente

On Tue, 27 Jan 2015, Chris Muller wrote:

This is easy for an expert to glance at and understand the nature of
the bug but could someone explain it to help our Smalltalk-user
community understand what has been wrong with MessageTally all these
years?

Thanks!

On Tue, Jan 27, 2015 at 12:54 PM, Bert Freudenberg <[hidden email]> wrote:

On 27.01.2015, at 19:37, Eliot Miranda <[hidden email]> wrote:


Wow, that's quite the bug :-)


Indeed. I wonder how this could have ever worked. Or maybe just nobody
noticed that certain chains were missing?

- Bert -



On Tue, Jan 27, 2015 at 8:53 AM, <[hidden email]> wrote:

Marcel Taeumel uploaded a new version of System to project The Inbox:
http://source.squeak.org/inbox/System-mt.697.mcz

==================== Summary ====================

Name: System-mt.697
Author: mt
Time: 27 January 2015, 5:53:38.431 pm
UUID: 9296f429-477d-0745-a878-6e11e7de5edd
Ancestors: System-cmm.696

Let message tally ignore #home but just use #sender to handle block
closures correctly.

=============== Diff against System-cmm.696 ===============

Item was changed:
  ----- Method: MessageTally>>tally:by: (in category 'tallying') -----
  tally: context by: count
        "Explicitly tally the specified context and its stack."
        | sender |

        "Add to this node if appropriate"
        context method == method ifTrue: [^self bumpBy: count].

        "No sender? Add new branch to the tree."
+       (sender :=  context sender) ifNil: [
-       (sender :=  context home sender)ifNil: [
                ^ (self bumpBy: count) tallyPath: context by: count].

        "Find the node for the sending context (or add it if necessary)"
        ^ (self tally: sender by: count) tallyPath: context by: count!

Item was changed:
  ----- Method: MessageTally>>tally:in:by: (in category 'tallying') -----
  tally: context in: aProcess by: count
        "Explicitly tally the specified context and its stack."
        | sender |

        "Add to this node if appropriate"
        context method == method ifTrue: [^self bumpBy: count].

        "No sender? Add new branch to the tree."
+       (sender :=  context sender) ifNil: [
-       (sender :=  context home sender) ifNil: [
                ^ (self bumpBy: count) tallyPath: context in: aProcess by:
count].

        "Find the node for the sending context (or add it if necessary)"
        ^ (self tally: sender in: aProcess by: count) tallyPath: context
in: aProcess by: count!

--
best,
Eliot

On Tue, Jan 27, 2015 at 04:14:24PM -0800, Eliot Miranda wrote:
> On Jan 27, 2015, at 4:00 PM, "David T. Lewis" <[hidden email]> wrote:
> > On Tue, Jan 27, 2015 at 01:51:52PM -0800, Eliot Miranda wrote:
> >> On Tue, Jan 27, 2015 at 1:39 PM, Tobias Pape <[hidden email]> wrote:
> >>> On 27.01.2015, at 22:21, Levente Uzonyi <[hidden email]> wrote:
> >>>
> >>>> AndreasSystemProfiler
> >>>
> >>> This is not in trunk, right?
> >>> If so, why?
> >>
> >> It came from Qwaq/Terf so it ended up in
> >> http://ss3.gemstone.com/ss/AndreasSystemProfiler.  Further, as Levente says
> >> it depends on support that is in Cog VMs (JIT, Stack & Interpreter) but not
> >> (yet) in the trunk Interpreter.  So I guess it doesn't belong in trunk
> >> until the Cog Interpreter has been merged with the trunk Interpreter.
> >> That's a fair amount of work, and neither David nor I have the time for it
> >> right now.
> >
> > I did spend some time last year trying to get the primitives working in
> > trunk Interpreter, but I was a bit out of my depth and did not get it right.
>
> Since the code is in the Cog Interpreter the right way to do this IMO us
> to port bug fixes to primitives etc into the Cog Interpreter and call the
> Cog VMMaker trunk.  Going the other way takes longer to get to the same place.
>

Hi Eliot,

Was Andreas' profiler fully functional on an Interpreter generated from oscog
at one point? I know that Andreas had done at least an initial implementation
in that part of the code. I had assumed that his main focus in that time frame
would have been on Cog, so I was not entirely sure if the Interpreter
implementation was complete.

Of course I was trying to adopt the implementation from the oscog branch
when I looked at this last year. But given that I was not successful, I
won't attempt to argue that this was the best possible approach.

Thanks,
Dave