Hi peter,

On Sat, 6 Jan 2001 08:06:23 -0500 "Peter Schuller" <peter.schuller at infidyne.com> wrote:
>But so there is absolutely no depdencent what so ever, except for the
>default name of the image it tries to load? I already know that the VM is
>more or less separate from the Smalltalk stuff (i.e. the image), but I was
>thinking there might be some details in the image low-level stuff that might
>be best left to the "right" VM.

The VMs generally work well from one release of the image to the next. An exception in recent memory was somewhere in 2.8 some code was changed relating to the lookup of external primitives, IIRC. At that point a new VM was strongly recommended to avoid decreased performance. There may come a time when, e.g. bytecodes change or the layout of a class the VM really depends on changes, that a new VM will be *required*, but that hasn't happened in a while.
 
>Another reason I preferred this way of updating is that I read a comment
>from someone (don't remember who - no offence :)) who had found that Morphic
>was faster with an image updates manually with the changes file rather than
>using the stock image (IIRC).
>
>Based on *very* simple tests I seem to be seeing the same thing. For
>example, scrolling the list of all methods of Object in the class browser in
>a morphic project leads to 100% CPU utilitzation in the 2.7 VM/stock 2.9
>image combination (on an Athlon 500, Linux). But doing the same thing with
>my new 'fresh' updates 2.8 image results in 50% CPU utilization.

It may be that something "happened" to the stock image. If someone who is experiencing this disparity of behavior were to run a MessageTally for 10 or 20 seconds in each of these images, it might help us figure out what the cause is.

Cheers,
Bob