Strange situation

[mark@cambridge.apple.com (Mark Preece)]

> There is a way to allocate more memory to the Mac heap before
>startup.  
>
>Go into ResEdit and open up MCL.
>Double click on LSIZ.
[and so on...]

>NOTE/WARNING:  The above is a true hack.  I have read no documentation and have
>no real knowlege in this are.  It does however seem to work.  
[...]

Hello Jeff--

We built a (slightly) more elegant interface to setting the Mac heap size
into MCL 2.0. It was too late to make it into the Reference Manual, but
it's in the MCL 2.0 Release Notes. Here's the story:

Save-Application

SAVE-APPLICATION takes a new keyword argument   :MEMORY-OPTIONS. This
argument gives you some control over how the saved application apportions
and uses the memory available to it.  The value of the :MEMORY-OPTIONS
keyword argument is itself a list of keyword/value pairs, which are
interpreted as follows:

                Keyword Default value
        :MAC-HEAP-MINIMUM       102400
        :MAC-HEAP-MAXIMUM       409600
        :MAC-HEAP-PERCENTAGE    5
        :STACK-MINIMUM  32768
        :STACK-MAXIMUM  184320 
        :STACK-PERCENTAGE       6
        :LOW-MEMORY-THRESHOLD   24576
        :COPYING-GC-THRESHOLD   2147483648

Given an application partition size of n bytes, the  MCL kernel reserves roughly

        (max
                mac-heap-minimum
                (min mac-heap-maximum (* n (/ mac-heap-percentage 100)))))

bytes for the "mac heap", and roughly

        (max
                stack-heap-minimum
                (min stack-heap-maximum (* n (/ stack-heap-percentage 100)))))

bytes for MCL's control and value stacks.  (Each stack uses roughly half of
this space.)
The MCL kernel signals an "OUT-OF-MEMORY" condition when, after GC, free
space in the lisp heap is less than the value of the :LOW-MEMORY-THRESHOLD
option.

If free space in the lisp heap exceeds the value of the
:COPYING-GC-THRESHOLD option when the application starts up, the MCL kernel
arranges that a two-space copying GC algorithm is used for (non-ephemeral)
garbage collection. This algorithm may improve GC performance in some large
VM environments, at the cost of roughly halving the amount of memory
available for the allocation of lisp objects.