[Coco] Original OS-9 Level 1 source code

[Neal Crook]

Hi Felipe,

Your email prompted me to look at the size of the files on my L1 system.
The platform is my multicomp09 FPGA system so the module set and file sizes
will obviously be different for a Coco or other platform.

The boot process involves a "kernelfile" loaded from track34 and a
"bootfile". Each of these files consists of a set of concatenated modules.
Memory is assigned in increments of 256 byte "pages". The concatenated
files are good and bad. Good because memory is assigned to each of these
two files "atomically" - there is no "alignment space" wasted between
modules as there would be if each module was loaded individually. Bad
because, once loaded, no module in either of these files can be removed
(the memory is permanently allocated).

[you know all this stuff already I expect, I'm just including it for
completeness and so that anyone can point out errors that I have made]

Here are the module sizes in bytes (decimal). No allowance is made for data
reservations.

The L1 kernelfile:

rel                    81
krn                  2041
krnp2                1355
init                  112
boot_sdc              240

The L1 bootfile:

ioman                1802
rbf                  3557
dds0_80d.dd            47
mc09sdc.dr            239
s0_80d.dd              53
s1_80d.dd              53
s2_80d.dd              53
s3_80d.dd              53
scf.mn               1747
mc6850.dr            1110
term_mc6850.dt         63
term_mc6850_t0.dt      61
term_mc6850_t1.dt      61
pipeman.mn            547
piper.dr               40
pipe.dd                38
mc09clock_50hz        347
clock2_soft           102
sysgo_dd              483

In order to save memory, you need to save increments of 256 byte "pages".

Obviously, the first thing to do is to shave down the kernelfile and
bootfile to remove any modules that are not strictly necessary.

There is no reason for the rel, init or boot_sdc modules to remain in
memory after boot (well, it should allow you to reboot your system without
doing a disk access but that is not a compelling reason). The current
design of L1 rel requires krn being straight after rel, but it would be
simple to remove that requirement. Then, the code that copies the
kernelfile to high memory could just copy krn, krnp2 and leave the other
stuff at $2600. That could simply be discarded after bootfile has been
loaded.All of that is quite simple but would only save 1-and-a-bit pages.

There's no shell in this list, and I think that contributes about 1.5K. It
should be possible to build a minimum-functionality command starter that is
smaller than that. Is it possible to run without
pipeman,mn/piper.dr/pipe.dd (another 625 bytes..).

The next thing to look at is whether there are modules that include any
one-time code setup. This could be moved to REL and discarded after boot.
Of course, that starts to compromise the modularity of the system in an
unappealing way.

I would be interested to see comparable module sizes for the original OS9
modules, and to hear about any other lean/mean approaches to reducing the
memory footprint

regards,

Neal.