MOL source tree moved


Subject: MOL source tree moved
From: Samuel Rydh (samuel@ibrium.se)
Date: Mon Mar 12 2001 - 19:03:16 MST


The MOL source tree has moved to a new host (both the rsync
and the bitkeeper server):

via BitKeeper:
        
        bk parent bk://zorn.theophys.kth.se:5000
        bk clone

vua rsync:

        rsync -vcaz zorn.theophys.kth.se::mol mol

Note that the identity of the repository has changed - it is
necessary to reclone the repository.

It should now be simple to build and install MOL from the
developer tree. It should be sufficient to do

        rsync -vcaz zorn.theophys.kth.se::mol mol
        make libimport
        make install

or, using bitkeeper,

        bk update
        make libimport
        make install

The 'libimport' rule downloads (using rsync) the support
binaries from zorn (the start sound, MacOS compiled drivers
and a few other things).

The latest development snapshot contains *extensive*
changes to the low-level parts of MOL. I'd like to get
some feedback before I roll it out as 0.9.56 (in particular
about performance and stability). The modifications in brief:

* Rewritten MOL engine:

    The source is now much cleaner and simpler (and slightly
    faster). Most assumptions about the kernel has been
    eliminated (the mol module should in principal be compatible
    with any 2.4 kernel without recompilation, although it
    is of course preferable to recompile it anyway).

    SMP should be a lot simpler to implement. In principal,
    what is missing is some locking (multi-session should
    also be almost working).

    The runtime patching (on 2.2 kernels) is now almost unnecessary
    (only a single hook remains).

    It should be a lot simpler to port MOL to APUS (as far as
    I'm aware, what needs to be fixed is APUS nonstandard
    KERNELBASE definition and 603 support).

* AltiVec support

    I've not tested this extensively since I didn't find
    many programs that actually used AltiVec...

* MacOS 9.1 stability fixes.

    I think I've addressed the problem which caused some setups
    to freeze.

* L2CR register is now handled correctly.

    Extensions setting the L2CR register should no longer
    freeze MOL (please confirm this).

To be fixed
===========

- MacOS 9.1 performance in MOL is slightly worse than 9.0.4.
I think this is caused by extensive table overflows in the
MMU subsystem. Fixing this should give a substantial performance
gain.

- The latest version of MOL will not run on 603 or 601 machines.
I will try to add true support for these machines (at least the 603)
later this week.

Cheers,

/Samuel

----------------------------------------------------------
 E-mail <samuel@ibrium.se> WWW: <http://www.ibrium.se>
  Phone/fax: (home) +46 8 4418431, (work) +46 8 7908470
----------------------------------------------------------



This archive was generated by hypermail 2a24 : Mon Mar 12 2001 - 19:05:27 MST