Tracing context switches

Hey all

what do you think is the best way to trace all context switches
during start up to the stdout?

Best regards
        -Tobias

Hi Tobias,

    IMO the best way would be to create a special VM that added printing to the transferTi:from: method (IIRC).  It can't be done reliably at the image level given the current implementation.  For example, you could add tracing to suspend, resume, signal and wait by renaming the primutives and invoking them from methods with the old name that add tracing, e.g.

Process>>resume
     self logResume.
      ^self primResume

but this a) isn't thread safe and b) misses context switches initiated by the VM, eg when a delay fires.

But doing it in the VM is safe and will catch all switches.  We can discuss further tomorrow.  It's a simple change and generating eg a new Stack VM should be straightforward.  We could think of adding this facility to the standard VM.

BTW, I'm glad you're looking at this.  I'm sure we can simplify start up a lot and hence speed up start up, which is really important eg for cloud computing on containers like docker.

_,,,^..^,,,_ (phone)

Hi,


On 08.10.2015, at 10:21, Eliot Miranda <[hidden email]> wrote:

> Hi Tobias,
>
>    IMO the best way would be to create a special VM that added printing to the transferTi:from: method (IIRC).  It can't be done reliably at the image level given the current implementation.  For example, you could add tracing to suspend, resume, signal and wait by renaming the primutives and invoking them from methods with the old name that add tracing, e.g.
>
> Process>>resume
>     self logResume.
>      ^self primResume
>
> but this a) isn't thread safe and b) misses context switches initiated by the VM, eg when a delay fires.

Yep, that's what I encountered, as in, the image didn't come up after such changes ;)


>
> But doing it in the VM is safe and will catch all switches.  We can discuss further tomorrow.  It's a simple change and generating eg a new Stack VM should be straightforward.  We could think of adding this facility to the standard VM.
>
> BTW, I'm glad you're looking at this.  I'm sure we can simplify start up a lot and hence speed up start up, which is really important eg for cloud computing on containers like docker.

Probably ;)
Currently I'm looking into this as we have problems with the networking code
on RSqueak VM and I suspect that there's just context switching gone wrong…

Best regards
        -Tobias